From 2817ade782cc13b8f42d9616e4b3a935400b5179 Mon Sep 17 00:00:00 2001
From: Omar Alvarez <omar.alvarez@cinfo.es>
Date: Mon, 22 Jun 2020 13:40:15 +0200
Subject: [PATCH 01/53] Fix parsing label files with unexpected chars

---
 src/Int8BatchStream.cpp | 19 ++++++-------------
 1 file changed, 6 insertions(+), 13 deletions(-)

diff --git a/src/Int8BatchStream.cpp b/src/Int8BatchStream.cpp
index dd4399f..fdc1db2 100644
--- a/src/Int8BatchStream.cpp
+++ b/src/Int8BatchStream.cpp
@@ -132,21 +132,14 @@ void BatchStream::readCVimage(std::string inputFileName, std::vector<float>& res
 
 void BatchStream::readLabels(std::string inputFileName, std::vector<float>& ris) {
     std::ifstream is(inputFileName.c_str());
-    //read only the first number: the image sub-portion class
-    while (true) {
+    
+    std::string line;
+    while (std::getline(is, line))
+    {
+        std::istringstream iss(line);
         float val;
-		is >> val;
-        if (!is) {
-            break;
-        }
-        // insert the first number and skip all others
+        if(!(iss >> val)) { break; } // error
         ris.push_back(val);
-        while( true ) {
-            char c;
-            is >> c;
-            if (is.peek() == '\n') //detect "\n"
-                break;
-        }   
     }
 }
 

From 61aa24c6b7716b833c8a0840dc0c47a00ca7e8ff Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Tue, 30 Jun 2020 15:19:03 +0200
Subject: [PATCH 02/53] yolov4tiny works on CUDNN

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 include/tkDNN/DarknetParser.h   |   1 +
 include/tkDNN/Layer.h           |   4 +-
 src/DarknetParser.cpp           |   4 +-
 src/Route.cpp                   |  10 +-
 tests/darknet/cfg/yolo4tiny.cfg | 281 ++++++++++++++++++++++++++++++++
 tests/darknet/yolo4tiny.cpp     |  33 ++++
 6 files changed, 328 insertions(+), 5 deletions(-)
 create mode 100644 tests/darknet/cfg/yolo4tiny.cfg
 create mode 100644 tests/darknet/yolo4tiny.cpp

diff --git a/include/tkDNN/DarknetParser.h b/include/tkDNN/DarknetParser.h
index f36469b..29d1e8e 100644
--- a/include/tkDNN/DarknetParser.h
+++ b/include/tkDNN/DarknetParser.h
@@ -11,6 +11,7 @@ namespace tk { namespace dnn {
         int channels = 3;
         int batch_normalize=0;
         int groups = 1;
+        int group_id = 0;
         int filters=1;
         int size_x=1;
         int size_y=1;
diff --git a/include/tkDNN/Layer.h b/include/tkDNN/Layer.h
index 9bd8432..bd544b2 100644
--- a/include/tkDNN/Layer.h
+++ b/include/tkDNN/Layer.h
@@ -509,7 +509,7 @@ public:
 class Route : public Layer {
 
 public:
-    Route(Network *net, Layer **layers, int layers_n); 
+    Route(Network *net, Layer **layers, int layers_n, int groups = 1, int group_id = 0); 
     virtual ~Route();
     virtual layerType_t getLayerType() { return LAYER_ROUTE; };
 
@@ -519,6 +519,8 @@ public:
     static const int MAX_LAYERS = 32;
     Layer *layers[MAX_LAYERS];  //ids of layers to be merged
     int layers_n; //number of layers
+    int groups;
+    int group_id;
 };
 
 
diff --git a/src/DarknetParser.cpp b/src/DarknetParser.cpp
index 5092595..7bbc7ba 100644
--- a/src/DarknetParser.cpp
+++ b/src/DarknetParser.cpp
@@ -75,6 +75,8 @@ namespace tk { namespace dnn {
             fields.coords = std::stoi(value);
         else if(name.find("groups") !=  std::string::npos)
             fields.groups = std::stoi(value);
+        else if(name.find("group_id") !=  std::string::npos)
+            fields.group_id = std::stoi(value);
         else if(name.find("scale_x_y") !=  std::string::npos)
             fields.scale_xy = std::stof(value);
         else if(name.find("from") !=  std::string::npos)
@@ -148,7 +150,7 @@ namespace tk { namespace dnn {
                 //std::cout<<"Route to "<<layerIdx<<" "<<netLayers[layerIdx]->getLayerName()<<"\n";
                 layers.push_back(netLayers[layerIdx]);
             }
-            netLayers.push_back(new tk::dnn::Route(net, layers.data(), layers.size()));
+            netLayers.push_back(new tk::dnn::Route(net, layers.data(), layers.size(), f.groups, f.group_id));
 
         } else if(f.type == "reorg") {
             netLayers.push_back(new tk::dnn::Reorg(net, f.stride_x));
diff --git a/src/Route.cpp b/src/Route.cpp
index 39bb14e..816566e 100644
--- a/src/Route.cpp
+++ b/src/Route.cpp
@@ -5,7 +5,7 @@
 
 namespace tk { namespace dnn {
 
-Route::Route(Network *net, Layer **layers, int layers_n) : Layer(net) {
+Route::Route(Network *net, Layer **layers, int layers_n, int groups, int group_id) : Layer(net) {
 
     // copy input layers
     if(layers_n > MAX_LAYERS) {
@@ -15,6 +15,8 @@ Route::Route(Network *net, Layer **layers, int layers_n) : Layer(net) {
         this->layers[i] = layers[i];
     }
     this->layers_n = layers_n;
+    this->groups = groups;
+    this->group_id = group_id;
     
     //get dims
     output_dim.l = 1;
@@ -32,6 +34,7 @@ Route::Route(Network *net, Layer **layers, int layers_n) : Layer(net) {
         output_dim.c += layers[i]->output_dim.c;
     }
 
+    output_dim.c /= this->groups;
     input_dim = output_dim;
 
     checkCuda( cudaMalloc(&dstData, output_dim.tot()*sizeof(dnnType)) );
@@ -49,8 +52,9 @@ dnnType* Route::infer(dataDim_t &dim, dnnType* srcData) {
     for(int i=0; i<layers_n; i++) {
         dnnType *input = layers[i]->dstData;
         int in_dim = layers[i]->output_dim.tot();
-        checkCuda( cudaMemcpy(dstData + offset, input, in_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice));
-        offset += in_dim;
+        int part_in_dim = in_dim / this->groups;
+        checkCuda( cudaMemcpy(dstData + offset, input + this->group_id*part_in_dim, part_in_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice));
+        offset += part_in_dim;
     }
 
     //update data dimensions    
diff --git a/tests/darknet/cfg/yolo4tiny.cfg b/tests/darknet/cfg/yolo4tiny.cfg
new file mode 100644
index 0000000..dc6f5bf
--- /dev/null
+++ b/tests/darknet/cfg/yolo4tiny.cfg
@@ -0,0 +1,281 @@
+[net]
+# Testing
+#batch=1
+#subdivisions=1
+# Training
+batch=64
+subdivisions=1
+width=416
+height=416
+channels=3
+momentum=0.9
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+learning_rate=0.00261
+burn_in=1000
+max_batches = 500200
+policy=steps
+steps=400000,450000
+scales=.1,.1
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=2
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers=-1
+groups=2
+group_id=1
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1,-2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -6,-1
+
+[maxpool]
+size=2
+stride=2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+##################################
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=linear
+
+
+
+[yolo]
+mask = 3,4,5
+anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
+classes=80
+num=6
+jitter=.3
+scale_x_y = 1.05
+cls_normalizer=1.0
+iou_normalizer=0.07
+iou_loss=ciou
+ignore_thresh = .7
+truth_thresh = 1
+random=0
+resize=1.5
+nms_kind=greedynms
+beta_nms=0.6
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[upsample]
+stride=2
+
+[route]
+layers = -1, 23
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=linear
+
+[yolo]
+mask = 1,2,3
+anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
+classes=80
+num=6
+jitter=.3
+scale_x_y = 1.05
+cls_normalizer=1.0
+iou_normalizer=0.07
+iou_loss=ciou
+ignore_thresh = .7
+truth_thresh = 1
+random=0
+resize=1.5
+nms_kind=greedynms
+beta_nms=0.6
diff --git a/tests/darknet/yolo4tiny.cpp b/tests/darknet/yolo4tiny.cpp
new file mode 100644
index 0000000..d9011a8
--- /dev/null
+++ b/tests/darknet/yolo4tiny.cpp
@@ -0,0 +1,33 @@
+#include<iostream>
+#include<vector>
+#include "tkdnn.h"
+#include "test.h"
+#include "DarknetParser.h"
+
+int main() {
+    std::string bin_path  = "yolo4tiny";
+    std::vector<std::string> input_bins = { 
+        bin_path + "/layers/input.bin"
+    };
+    std::vector<std::string> output_bins = {
+        bin_path + "/debug/layer30_out.bin",
+        bin_path + "/debug/layer37_out.bin"
+    };
+    std::string wgs_path  = bin_path + "/layers";
+    std::string cfg_path  = std::string(TKDNN_PATH) + "/tests/darknet/cfg/yolo4tiny.cfg";
+    std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/coco.names";
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/iRnc4pSqmx78gJs/download");
+
+    // parse darknet network
+    tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
+    net->print();
+
+    //convert network to tensorRT
+    // tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
+    
+    int ret = testInference(input_bins, output_bins, net, nullptr);
+    net->releaseLayers();
+    delete net;
+    // delete netRT;
+    return ret;
+}

From fe2e4eae92324d4c03482e20b5746ebc6aed75c5 Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Tue, 30 Jun 2020 15:52:58 +0200
Subject: [PATCH 03/53] yolo4tiny works on tensorRT :dolphin: :dolphin:
 :dolphin: :dolphin:

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 include/tkDNN/NetworkRT.h         |  2 +-
 include/tkDNN/pluginsRT/RouteRT.h | 17 ++++++++++++-----
 src/NetworkRT.cpp                 | 19 +++++++++++--------
 tests/darknet/yolo4tiny.cpp       |  6 +++---
 4 files changed, 27 insertions(+), 17 deletions(-)

diff --git a/include/tkDNN/NetworkRT.h b/include/tkDNN/NetworkRT.h
index ee1f728..66b4f3d 100644
--- a/include/tkDNN/NetworkRT.h
+++ b/include/tkDNN/NetworkRT.h
@@ -28,7 +28,7 @@ using namespace nvinfer1;
 #include "pluginsRT/ActivationMishRT.h"
 #include "pluginsRT/ReorgRT.h"
 #include "pluginsRT/RegionRT.h"
-//#include "pluginsRT/RouteRT.h"
+#include "pluginsRT/RouteRT.h"
 #include "pluginsRT/ShortcutRT.h"
 #include "pluginsRT/YoloRT.h"
 #include "pluginsRT/UpsampleRT.h"
diff --git a/include/tkDNN/pluginsRT/RouteRT.h b/include/tkDNN/pluginsRT/RouteRT.h
index 0e94a97..263893f 100644
--- a/include/tkDNN/pluginsRT/RouteRT.h
+++ b/include/tkDNN/pluginsRT/RouteRT.h
@@ -8,7 +8,9 @@ class RouteRT : public IPlugin {
 	*/
 
 public:
-	RouteRT() {
+	RouteRT(int groups, int group_id) {
+		this->groups = groups;
+		this->group_id = group_id;
 	}
 
 	~RouteRT(){
@@ -22,7 +24,7 @@ public:
 	Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) override {
 		int out_c = 0;
 		for(int i=0; i<nbInputDims; i++) out_c += inputs[i].d[0];
-		return DimsCHW{out_c, inputs[0].d[1], inputs[0].d[2]};
+		return DimsCHW{out_c/groups, inputs[0].d[1], inputs[0].d[2]};
 	}
 
 	void configure(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, int maxBatchSize) override {
@@ -34,6 +36,7 @@ public:
 		}
 		h = inputDims[0].d[1];
 		w = inputDims[0].d[2];
+		c /= groups;
 	}
 
 	int initialize() override {
@@ -56,8 +59,9 @@ public:
 		for(int i=0; i<in; i++) {
 			dnnType *input = (dnnType*)reinterpret_cast<const dnnType*>(inputs[i]);
 			int in_dim = c_in[i]*h*w;
-			checkCuda( cudaMemcpyAsync(dstData + offset, input, in_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice, stream) );
-			offset += in_dim;
+			int part_in_dim = in_dim / this->groups;
+			checkCuda( cudaMemcpyAsync(dstData + offset, input + this->group_id*part_in_dim, part_in_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice, stream) );
+			offset += part_in_dim;
 		}
 
 		return 0;
@@ -65,11 +69,13 @@ public:
 
 
 	virtual size_t getSerializationSize() override {
-		return (4+MAX_INPUTS)*sizeof(int);
+		return (6+MAX_INPUTS)*sizeof(int);
 	}
 
 	virtual void serialize(void* buffer) override {
 		char *buf = reinterpret_cast<char*>(buffer);
+		tk::dnn::writeBUF(buf, groups);
+		tk::dnn::writeBUF(buf, group_id);
 		tk::dnn::writeBUF(buf, in);
 		for(int i=0; i<MAX_INPUTS; i++)
 			tk::dnn::writeBUF(buf, c_in[i]);
@@ -83,4 +89,5 @@ public:
 	int in;
 	int c_in[MAX_INPUTS];
 	int c, h, w;
+	int groups, group_id;
 };
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index 6e86de1..a6725aa 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -449,12 +449,15 @@ ILayer* NetworkRT::convert_layer(ITensor *input, Route *l) {
         // }
         // std::cout<<"\n";
     }
-    
-    IConcatenationLayer *lRT = networkRT->addConcatenation(tens, l->layers_n);
-    //IPlugin *plugin = new RouteRT();
-    //IPluginLayer *lRT = networkRT->addPlugin(tens, l->layers_n, *plugin);
-    checkNULL(lRT);
 
+    if(l->groups > 1){
+        IPlugin *plugin = new RouteRT(l->groups, l->group_id);
+        IPluginLayer *lRT = networkRT->addPlugin(tens, l->layers_n, *plugin);
+        checkNULL(lRT);
+        return lRT;
+    }
+    IConcatenationLayer *lRT = networkRT->addConcatenation(tens, l->layers_n);
+    checkNULL(lRT);
     return lRT;
 }
 
@@ -766,9 +769,9 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         r->w = readBUF<int>(buf);
         return r;
     } 
-/*
+
     if(name.find("Route") == 0) {
-        RouteRT *r = new RouteRT();
+        RouteRT *r = new RouteRT(readBUF<int>(buf),readBUF<int>(buf));
         r->in = readBUF<int>(buf);
         for(int i=0; i<RouteRT::MAX_INPUTS; i++)
             r->c_in[i] = readBUF<int>(buf);
@@ -777,7 +780,7 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         r->w = readBUF<int>(buf);
         return r;
     } 
-*/
+
     if(name.find("Deformable") == 0) {
         DeformableConvRT *r = new DeformableConvRT(readBUF<int>(buf), readBUF<int>(buf), readBUF<int>(buf),
                                                     readBUF<int>(buf), readBUF<int>(buf), readBUF<int>(buf),
diff --git a/tests/darknet/yolo4tiny.cpp b/tests/darknet/yolo4tiny.cpp
index d9011a8..44fbac8 100644
--- a/tests/darknet/yolo4tiny.cpp
+++ b/tests/darknet/yolo4tiny.cpp
@@ -23,11 +23,11 @@ int main() {
     net->print();
 
     //convert network to tensorRT
-    // tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
+    tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
     
-    int ret = testInference(input_bins, output_bins, net, nullptr);
+    int ret = testInference(input_bins, output_bins, net, netRT);
     net->releaseLayers();
     delete net;
-    // delete netRT;
+    delete netRT;
     return ret;
 }

From 04602f395236d993d79cff7d1be5ced3ae738b0b Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Tue, 30 Jun 2020 19:37:16 +0200
Subject: [PATCH 04/53] Yolo4-tiny batched fix #59

---
 README.md                         |  2 ++
 include/tkDNN/pluginsRT/RouteRT.h | 18 ++++++++++--------
 scripts/test_all_tests.sh         |  1 +
 3 files changed, 13 insertions(+), 8 deletions(-)

diff --git a/README.md b/README.md
index 19b98a2..3ff6fe7 100644
--- a/README.md
+++ b/README.md
@@ -317,6 +317,8 @@ This demo also creates a json file named ```net_name_COCO_res.json``` containing
 | resnet101_cnet    | Centernet (Resnet101 backend)<sup>4</sup>     | [COCO 2017](http://cocodataset.org/)                          | 80        | 512x512       | [weights](https://cloud.hipert.unimore.it/s/5BTjHMWBcJk8g3i/download)     |
 | csresnext50-panet-spp    | Cross Stage Partial Network <sup>7</sup>     | [COCO 2014](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/Kcs4xBozwY4wFx8/download)     |
 | yolo4             | Yolov4 <sup>8</sup>                           | [COCO 2017](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/d97CFzYqCPCp5Hg/download)     |
+| yolo4_berkeley             | Yolov4 <sup>8</sup>                           | [BDD100K  ](https://bair.berkeley.edu/blog/2018/05/30/bdd/)                          | 10        | 540x320       | [weights](https://cloud.hipert.unimore.it/s/nkWFa5fgb4NTdnB/download)     |
+| yolo4tiny             | Yolov4 tiny                           | [COCO 2017](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/iRnc4pSqmx78gJs/download)     |
 
 
 ## References
diff --git a/include/tkDNN/pluginsRT/RouteRT.h b/include/tkDNN/pluginsRT/RouteRT.h
index 263893f..23f30b7 100644
--- a/include/tkDNN/pluginsRT/RouteRT.h
+++ b/include/tkDNN/pluginsRT/RouteRT.h
@@ -52,16 +52,18 @@ public:
 	}
 
 	virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override {
-
+		
 		dnnType *dstData = reinterpret_cast<dnnType*>(outputs[0]);
 
-		int offset = 0;
-		for(int i=0; i<in; i++) {
-			dnnType *input = (dnnType*)reinterpret_cast<const dnnType*>(inputs[i]);
-			int in_dim = c_in[i]*h*w;
-			int part_in_dim = in_dim / this->groups;
-			checkCuda( cudaMemcpyAsync(dstData + offset, input + this->group_id*part_in_dim, part_in_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice, stream) );
-			offset += part_in_dim;
+		for(int b=0; b<batchSize; b++) {
+			int offset = 0;
+			for(int i=0; i<in; i++) {
+				dnnType *input = (dnnType*)reinterpret_cast<const dnnType*>(inputs[i]);
+				int in_dim = c_in[i]*h*w;
+				int part_in_dim = in_dim / this->groups;
+				checkCuda( cudaMemcpyAsync(dstData + b*c*w*h + offset, input + b*c*w*h*groups + this->group_id*part_in_dim, part_in_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice, stream) );
+				offset += part_in_dim;
+			}
 		}
 
 		return 0;
diff --git a/scripts/test_all_tests.sh b/scripts/test_all_tests.sh
index 193a90c..770aa22 100644
--- a/scripts/test_all_tests.sh
+++ b/scripts/test_all_tests.sh
@@ -74,6 +74,7 @@ do
 
     test_net yolo4
     test_net yolo4_berkeley
+    test_net yolo4tiny
     test_net yolo3
     test_net yolo3_berkeley
     test_net yolo3_coco4

From 7c2155decfc2d225f523350d55fb6d773d3b3b6c Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Wed, 1 Jul 2020 11:56:46 +0200
Subject: [PATCH 05/53] Add weights download link for
 csresnext50-panet-spp_berkeley ( fix #63 )

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 tests/darknet/csresnext50-panet-spp_berkeley.cpp | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/tests/darknet/csresnext50-panet-spp_berkeley.cpp b/tests/darknet/csresnext50-panet-spp_berkeley.cpp
index 3cd0d52..a8ba59f 100644
--- a/tests/darknet/csresnext50-panet-spp_berkeley.cpp
+++ b/tests/darknet/csresnext50-panet-spp_berkeley.cpp
@@ -17,8 +17,7 @@ int main() {
     std::string wgs_path  = bin_path + "/layers";
     std::string cfg_path  = std::string(TKDNN_PATH) + "/tests/darknet/cfg/csresnext50-panet-spp_berkeley.cfg";
     std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/berkeley.names";
-    // FIXME: wrong weights
-    // downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s//download");
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/q82qHAtqpoaFYo5/download");
 
     // parse darknet network
     tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);

From b12cf0d7c2599f36e7e564ba19868c6885ecca6d Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Mon, 13 Jul 2020 19:50:00 +0200
Subject: [PATCH 06/53] docker

---
 docker/Dockerfile      |  7 ++++++
 docker/Dockerfile.base | 57 ++++++++++++++++++++++++++++++++++++++++++
 docker/README.md       | 21 ++++++++++++++++
 3 files changed, 85 insertions(+)
 create mode 100644 docker/Dockerfile
 create mode 100644 docker/Dockerfile.base
 create mode 100644 docker/README.md

diff --git a/docker/Dockerfile b/docker/Dockerfile
new file mode 100644
index 0000000..3c9fb61
--- /dev/null
+++ b/docker/Dockerfile
@@ -0,0 +1,7 @@
+FROM ceccocats/tkdnn:latest
+LABEL maintainer "Francesco Gatti"
+
+RUN cd && git clone https://github.com/ceccocats/tkDNN.git && cd tkDNN && mkdir build && cd build \ 
+    && cmake .. && make -j12
+
+
diff --git a/docker/Dockerfile.base b/docker/Dockerfile.base
new file mode 100644
index 0000000..e61b0d3
--- /dev/null
+++ b/docker/Dockerfile.base
@@ -0,0 +1,57 @@
+FROM nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04
+LABEL maintainer "Francesco Gatti"
+
+ADD nv-tensorrt-repo-ubuntu1804-cuda10.2-trt7.0.0.11-ga-20191216_1-1_amd64.deb /tmp/trt.deb
+RUN apt-get update && dpkg -i /tmp/trt.deb && rm /tmp/trt.deb && apt-get update
+RUN apt install -y libnvinfer7=7.0.0-1+cuda10.2 libnvinfer-dev=7.0.0-1+cuda10.2
+RUN DEBIAN_FRONTEND=noninteractive apt install -y git wget libeigen3-dev libyaml-cpp-dev
+RUN cd /tmp && \
+    wget https://github.com/Kitware/CMake/releases/download/v3.17.3/cmake-3.17.3-Linux-x86_64.sh && \
+    chmod +x cmake-3.17.3-Linux-x86_64.sh && \
+    ./cmake-3.17.3-Linux-x86_64.sh --prefix=/usr/local --exclude-subdir --skip-license && \
+    rm ./cmake-3.17.3-Linux-x86_64.sh
+
+RUN echo "INSTALL OPENCV"
+RUN apt-get install -y build-essential \
+    unzip \
+    pkg-config \
+    libjpeg-dev \
+    libpng-dev \
+    libtiff-dev \
+    libavcodec-dev \
+    libavformat-dev \
+    libswscale-dev \
+    libv4l-dev \
+    libxvidcore-dev \
+    libx264-dev \
+    libgtk-3-dev \
+    libatlas-base-dev \
+    gfortran \
+    libgstreamer1.0-dev \
+    libgstreamer-plugins-base1.0-dev \
+    libdc1394-22-dev \
+    libavresample-dev
+RUN cd && wget https://github.com/opencv/opencv/archive/4.3.0.tar.gz && tar -xf 4.3.0.tar.gz && rm *.tar.gz
+RUN cd && wget https://github.com/opencv/opencv_contrib/archive/4.3.0.tar.gz && tar -xf 4.3.0.tar.gz && rm *.tar.gz
+RUN cd && \
+    cd opencv-4.3.0 && mkdir build && cd build && \
+    cmake -D CMAKE_BUILD_TYPE=RELEASE \
+    -D CMAKE_INSTALL_PREFIX=/usr/local \
+    -D INSTALL_PYTHON_EXAMPLES=OFF \
+    -D INSTALL_C_EXAMPLES=OFF \
+    -D OPENCV_EXTRA_MODULES_PATH='~/opencv_contrib-4.3.0/modules' \
+    -D BUILD_EXAMPLES=OFF \
+    -D WITH_CUDA=ON \
+    -D CUDA_ARCH_BIN=7.2 \
+    -D CUDA_ARCH_PTX="" \
+    -D ENABLE_FAST_MATH=ON \
+    -D CUDA_FAST_MATH=ON \
+    -D WITH_CUBLAS=ON \
+    -D WITH_LIBV4L=ON \
+    -D WITH_GSTREAMER=ON \
+    -D WITH_GSTREAMER_0_10=OFF \
+    -D WITH_TBB=ON \
+    ../ && make -j12 && make install
+RUN apt clean
+
+
diff --git a/docker/README.md b/docker/README.md
new file mode 100644
index 0000000..a3edf3c
--- /dev/null
+++ b/docker/README.md
@@ -0,0 +1,21 @@
+# Use the prebuilt image
+```
+# build image
+docker build -t tkdnn:build -f Dockerfile-f Dockerfile . 
+```
+
+# Build Base Docker image
+```
+# make nvidia docker working
+# follow this guide: https://github.com/NVIDIA/nvidia-docker
+
+# dowload tensorrt
+# from: https://developer.nvidia.com/compute/machine-learning/tensorrt/secure/7.0/7.0.0.11/local_repo/nv-tensorrt-repo-ubuntu1804-cuda10.2-trt7.0.0.11-ga-20191216_1-1_amd64.deb
+
+# build image
+docker build -t ceccocats/tkdnn:latest -f Dockerfile.base .
+
+# run image
+docker run -ti --gpus all --rm ceccocats/tkdnn:latest bash
+```
+

From b2df9fc1107ca63301df42a0fbd0483189eda7d9 Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Mon, 13 Jul 2020 19:51:09 +0200
Subject: [PATCH 07/53] Update README.md

---
 docker/README.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docker/README.md b/docker/README.md
index a3edf3c..aec202a 100644
--- a/docker/README.md
+++ b/docker/README.md
@@ -1,7 +1,7 @@
 # Use the prebuilt image
 ```
 # build image
-docker build -t tkdnn:build -f Dockerfile-f Dockerfile . 
+docker build -t tkdnn:build -f Dockerfile . 
 ```
 
 # Build Base Docker image

From c4aad7fe95e0f8ed45fd175d565d1926d2880f6c Mon Sep 17 00:00:00 2001
From: tk <micaelaverucchi@gmail.com>
Date: Thu, 16 Jul 2020 18:16:09 +0200
Subject: [PATCH 08/53] Patch for CUDNN 8.0.1

Signed-off-by: tk <micaelaverucchi@gmail.com>
---
 include/tkDNN/Layer.h |  4 ++--
 src/Conv2d.cpp        | 26 +++++++++++++-------------
 2 files changed, 15 insertions(+), 15 deletions(-)

diff --git a/include/tkDNN/Layer.h b/include/tkDNN/Layer.h
index bd544b2..f2ec56d 100644
--- a/include/tkDNN/Layer.h
+++ b/include/tkDNN/Layer.h
@@ -273,8 +273,8 @@ public:
 protected:
     cudnnFilterDescriptor_t filterDesc;
     cudnnConvolutionDescriptor_t convDesc;
-    cudnnConvolutionFwdAlgo_t     algo;
-    cudnnConvolutionBwdDataAlgo_t bwAlgo;
+    cudnnConvolutionFwdAlgoPerf_t     algo;
+    cudnnConvolutionBwdDataAlgoPerf_t bwAlgo;
     cudnnTensorDescriptor_t biasTensorDesc;
 
     void initCUDNN(bool back = false);
diff --git a/src/Conv2d.cpp b/src/Conv2d.cpp
index 4704c66..595fec7 100644
--- a/src/Conv2d.cpp
+++ b/src/Conv2d.cpp
@@ -63,23 +63,23 @@ void Conv2d::initCUDNN(bool back) {
     workSpace = NULL;
     ws_sizeInBytes = 0;
     if(back) {
-        checkCUDNN( cudnnGetConvolutionBackwardDataAlgorithm(net->cudnnHandle,
-                                                        filterDesc, dstTensor, convDesc, srcTensor,
-                                                        CUDNN_CONVOLUTION_BWD_DATA_PREFER_FASTEST, 0, &bwAlgo) );
+        checkCUDNN( cudnnGetConvolutionBackwardDataAlgorithm_v7(net->cudnnHandle,
+                                                                filterDesc, dstTensor, convDesc, srcTensor, 1, 0, &bwAlgo) );
         checkCUDNN(cudnnGetConvolutionBackwardDataWorkspaceSize(net->cudnnHandle,
-                                                           filterDesc, dstTensor, convDesc, srcTensor,
-                                                           bwAlgo, &ws_sizeInBytes));
+                                                                filterDesc, dstTensor, convDesc, srcTensor,
+                                                                bwAlgo.algo, &ws_sizeInBytes));
+        
 
         // invert tensors
         srcTensorDesc = dstTensor;
         dstTensorDesc = srcTensor;
     } else {
-        checkCUDNN( cudnnGetConvolutionForwardAlgorithm(net->cudnnHandle,
-                                                        srcTensor, filterDesc, convDesc, dstTensor,
-                                                        CUDNN_CONVOLUTION_FWD_PREFER_FASTEST, 0, &algo) );
-        checkCUDNN(cudnnGetConvolutionForwardWorkspaceSize(net->cudnnHandle,
-                                                           srcTensor, filterDesc, convDesc, dstTensor,
-                                                           algo, &ws_sizeInBytes));
+        checkCUDNN( cudnnGetConvolutionForwardAlgorithm_v7(net->cudnnHandle,
+                                                         srcTensor, filterDesc, convDesc, dstTensor,
+                                                        1, 0, &algo) );
+         checkCUDNN(cudnnGetConvolutionForwardWorkspaceSize(net->cudnnHandle,
+                                                            srcTensor, filterDesc, convDesc, dstTensor,
+                                                           algo.algo, &ws_sizeInBytes));
     }
 }
 
@@ -91,12 +91,12 @@ void Conv2d::inferCUDNN(dnnType* srcData, bool back) {
         checkCUDNN(cudnnConvolutionBackwardData(net->cudnnHandle,
                                                 &alpha, filterDesc, data_d,
                                                 srcTensorDesc, srcData,
-                                                convDesc, bwAlgo, workSpace, ws_sizeInBytes,
+                                                convDesc, bwAlgo.algo, workSpace, ws_sizeInBytes,
                                                 &beta, dstTensorDesc, dstData));
     } else {
         checkCUDNN(cudnnConvolutionForward(net->cudnnHandle,
                                            &alpha, srcTensorDesc, srcData, filterDesc,
-                                           data_d, convDesc, algo, workSpace, ws_sizeInBytes,
+                                           data_d, convDesc, algo.algo, workSpace, ws_sizeInBytes,
                                            &beta, dstTensorDesc, dstData));
     }
 

From 6a68f19b2ceb61542fe87533fb193f696d30c664 Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Thu, 16 Jul 2020 18:37:37 +0200
Subject: [PATCH 09/53] Fix patch from @ahmedius2 , tkDNN now supports CUDNN
 8.0.1 (Fix #74)

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
	       Francesco Gatti <gattifrancesco@hotmail.it>
---
 src/Conv2d.cpp | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/src/Conv2d.cpp b/src/Conv2d.cpp
index 595fec7..b57cf58 100644
--- a/src/Conv2d.cpp
+++ b/src/Conv2d.cpp
@@ -62,9 +62,10 @@ void Conv2d::initCUDNN(bool back) {
     // init workspace
     workSpace = NULL;
     ws_sizeInBytes = 0;
+    int algo_count = 0;
     if(back) {
         checkCUDNN( cudnnGetConvolutionBackwardDataAlgorithm_v7(net->cudnnHandle,
-                                                                filterDesc, dstTensor, convDesc, srcTensor, 1, 0, &bwAlgo) );
+                                                                filterDesc, dstTensor, convDesc, srcTensor, 1, &algo_count, &bwAlgo) );
         checkCUDNN(cudnnGetConvolutionBackwardDataWorkspaceSize(net->cudnnHandle,
                                                                 filterDesc, dstTensor, convDesc, srcTensor,
                                                                 bwAlgo.algo, &ws_sizeInBytes));
@@ -74,13 +75,17 @@ void Conv2d::initCUDNN(bool back) {
         srcTensorDesc = dstTensor;
         dstTensorDesc = srcTensor;
     } else {
+        
         checkCUDNN( cudnnGetConvolutionForwardAlgorithm_v7(net->cudnnHandle,
                                                          srcTensor, filterDesc, convDesc, dstTensor,
-                                                        1, 0, &algo) );
+                                                        1, &algo_count, &algo) );
          checkCUDNN(cudnnGetConvolutionForwardWorkspaceSize(net->cudnnHandle,
                                                             srcTensor, filterDesc, convDesc, dstTensor,
                                                            algo.algo, &ws_sizeInBytes));
     }
+
+    if(algo_count < 1)
+        FatalError("Cannot retrieve convolutional algo");
 }
 
 void Conv2d::inferCUDNN(dnnType* srcData, bool back) {

From f4970d1e6faab505c2caf1d6833cf7490a971a0e Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Fri, 17 Jul 2020 14:37:10 +0200
Subject: [PATCH 10/53] Update README.md

---
 README.md | 27 +++++++++++++++++++--------
 1 file changed, 19 insertions(+), 8 deletions(-)

diff --git a/README.md b/README.md
index 3ff6fe7..a1b5b16 100644
--- a/README.md
+++ b/README.md
@@ -1,9 +1,9 @@
 # tkDNN
-tkDNN is a Deep Neural Network library built with cuDNN and tensorRT primitives, specifically thought to work on NVIDIA Jetson Boards. It has been tested on TK1(branch cudnn2), TX1, TX2, AGX Xavier and several discrete GPU.
+tkDNN is a Deep Neural Network library built with cuDNN and tensorRT primitives, specifically thought to work on NVIDIA Jetson Boards. It has been tested on TK1(branch cudnn2), TX1, TX2, AGX Xavier, Nano and several discrete GPUs.
 The main goal of this project is to exploit NVIDIA boards as much as possible to obtain the best inference performance. It does not allow training. 
 
 
-If you use tkDNN in your research, please cite one of the following papers. For use in commercial solutions, write at gattifrancesco@hotmail.it or refer to https://hipert.unimore.it/ .
+If you use tkDNN in your research, please cite one of the following papers. For use in commercial solutions, write at gattifrancesco@hotmail.it and micaela.verucchi@unimore.it or refer to https://hipert.unimore.it/ .
 
 ```
 Accepted paper @ IRC 2020, will soon be published.
@@ -175,15 +175,25 @@ All models from darknet are now parsed directly from cfg, you still need to expo
   mish
 </details>
 
-## Run the demo
+## Run the demo 
+This is an example using yolov4.
 
-To run the an object detection demo follow these steps (example with yolov3):
+To run the an object detection first create the .rt file by running:
 ```
-rm yolo3_fp32.rt        # be sure to delete(or move) old tensorRT files
-./test_yolo3            # run the yolo test (is slow)
-./demo yolo3_fp32.rt ../demo/yolo_test.mp4 y
+rm yolo4_fp32.rt        # be sure to delete(or move) old tensorRT files
+./test_yolo4            # run the yolo test (is slow)
 ```
-In general the demo program takes 4 parameters:
+If you get problems in the creation, try to check the error activating the debug of TensorRT in this way:
+```
+cmake .. -DDEBUG=True
+make
+```
+
+Once you have succesfully created your rt file, run the demo: 
+```
+./demo yolo4_fp32.rt ../demo/yolo_test.mp4 y
+```
+In general the demo program takes 6 parameters:
 ```
 ./demo <network-rt-file> <path-to-video> <kind-of-network> <number-of-classes> <n-batches> <show-flag>
 ```
@@ -197,6 +207,7 @@ where
 
 N.b. By default it is used FP32 inference
 
+
 ![demo](https://user-images.githubusercontent.com/11562617/72547657-540e7800-388d-11ea-83c6-49dfea2a0607.gif)
 
 ### FP16 inference

From 3a0802d70c7a6286ac1ca871efc2498550295aa5 Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Mon, 27 Jul 2020 13:45:42 +0200
Subject: [PATCH 11/53] Resolve detection objects pick by prob threshold.
 Before this it will only pick the last object with prob > thresh wich is
 absolutely wrong Now it picks all the objects with prob > thesh.

fixes #94
---
 src/Yolo3Detection.cpp | 47 +++++++++++++++++++++---------------------
 1 file changed, 24 insertions(+), 23 deletions(-)

diff --git a/src/Yolo3Detection.cpp b/src/Yolo3Detection.cpp
index c76af20..27f393f 100644
--- a/src/Yolo3Detection.cpp
+++ b/src/Yolo3Detection.cpp
@@ -113,34 +113,35 @@ void Yolo3Detection::postprocess(const int bi, const bool mAP){
         int x1   = (b.x+b.w/2.);
         int y0   = (b.y-b.h/2.);
         int y1   = (b.y+b.h/2.);
-        int obj_class = -1;
-        float prob = 0;
+
         for(int c=0; c<classes; c++) {
             if(dets[j].prob[c] >= confThreshold) {
-                obj_class = c;
-                prob = dets[j].prob[c];
+                int obj_class = c;
+                float prob = dets[j].prob[c];
+
+                // convert to image coords
+                x0 = x_ratio*x0;
+                x1 = x_ratio*x1;
+                y0 = y_ratio*y0;
+                y1 = y_ratio*y1;
+                
+                tk::dnn::box res;
+                res.cl = obj_class;
+                res.prob = prob;
+                res.x = x0;
+                res.y = y0;
+                res.w = x1 - x0;
+                res.h = y1 - y0;
+
+                // FIXME: this shuld be useless
+                // if(mAP)
+                //     for(int c=0; c<classes; c++) 
+                //         res.probs.push_back(dets[j].prob[c]);
+
+                detected.push_back(res);
             }
         }
 
-        if(obj_class >= 0) {
-            // convert to image coords
-            x0 = x_ratio*x0;
-            x1 = x_ratio*x1;
-            y0 = y_ratio*y0;
-            y1 = y_ratio*y1;
-              
-            tk::dnn::box res;
-            res.cl = obj_class;
-            res.prob = prob;
-            res.x = x0;
-            res.y = y0;
-            res.w = x1 - x0;
-            res.h = y1 - y0;
-            if(mAP)
-                for(int c=0; c<classes; c++) 
-                    res.probs.push_back(dets[j].prob[c]);
-            detected.push_back(res);
-        }
     }
     batchDetected.push_back(detected);
 }

From a5d2d4792abe932131230b1d163d408092188b63 Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Mon, 27 Jul 2020 13:52:39 +0200
Subject: [PATCH 12/53] fix coords convert

---
 src/Yolo3Detection.cpp | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/src/Yolo3Detection.cpp b/src/Yolo3Detection.cpp
index 27f393f..606c6d1 100644
--- a/src/Yolo3Detection.cpp
+++ b/src/Yolo3Detection.cpp
@@ -114,17 +114,17 @@ void Yolo3Detection::postprocess(const int bi, const bool mAP){
         int y0   = (b.y-b.h/2.);
         int y1   = (b.y+b.h/2.);
 
+        // convert to image coords
+        x0 = x_ratio*x0;
+        x1 = x_ratio*x1;
+        y0 = y_ratio*y0;
+        y1 = y_ratio*y1;
+        
         for(int c=0; c<classes; c++) {
             if(dets[j].prob[c] >= confThreshold) {
                 int obj_class = c;
                 float prob = dets[j].prob[c];
 
-                // convert to image coords
-                x0 = x_ratio*x0;
-                x1 = x_ratio*x1;
-                y0 = y_ratio*y0;
-                y1 = y_ratio*y1;
-                
                 tk::dnn::box res;
                 res.cl = obj_class;
                 res.prob = prob;

From f778e1aa998f894654b24c0ab9ad759c0eb14019 Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Wed, 5 Aug 2020 19:55:10 +0200
Subject: [PATCH 13/53] Fixed boxes to float, add conf thresh as param

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 README.md                          |  3 ++-
 demo/config.yaml                   |  2 +-
 demo/demo/demo.cpp                 |  5 ++++-
 demo/demo/map.cpp                  |  2 +-
 include/tkDNN/CenternetDetection.h |  2 +-
 include/tkDNN/DetectionNN.h        |  2 +-
 include/tkDNN/MobilenetDetection.h |  2 +-
 include/tkDNN/Yolo3Detection.h     |  2 +-
 src/CenternetDetection.cpp         |  3 ++-
 src/MobilenetDetection.cpp         |  3 ++-
 src/Yolo3Detection.cpp             | 11 ++++++-----
 11 files changed, 22 insertions(+), 15 deletions(-)

diff --git a/README.md b/README.md
index a1b5b16..d9b5755 100644
--- a/README.md
+++ b/README.md
@@ -193,7 +193,7 @@ Once you have succesfully created your rt file, run the demo:
 ```
 ./demo yolo4_fp32.rt ../demo/yolo_test.mp4 y
 ```
-In general the demo program takes 6 parameters:
+In general the demo program takes 7 parameters:
 ```
 ./demo <network-rt-file> <path-to-video> <kind-of-network> <number-of-classes> <n-batches> <show-flag>
 ```
@@ -204,6 +204,7 @@ where
 *  ```<number-of-classes>```is the number of classes the network is trained on
 *  ```<n-batches>``` number of batches to use in inference (N.B. you should first export TKDNN_BATCHSIZE to the required n_batches and create again the rt file for the network).
 *  ```<show-flag>``` if set to 0 the demo will not show the visualization but save the video into result.mp4 (if n-batches ==1)
+*  ```<conf-thresh>``` confidence threshold for the detector. Only bounding boxes with threshold greater than conf-thresh will be displayed.
 
 N.b. By default it is used FP32 inference
 
diff --git a/demo/config.yaml b/demo/config.yaml
index e6f91a7..31ac599 100644
--- a/demo/config.yaml
+++ b/demo/config.yaml
@@ -3,5 +3,5 @@ map_points  : 101   #number of recall points (0 for all, 101 for COCO, 11 Pascal
 map_levels  : 10    #number of IoU step for the AP
 map_step    : 0.05  #step of IoU
 IoU_thresh  : 0.5   #starting IoU threshold
-conf_thresh : 0.0   #threshold on the condifence of the bbox
+conf_thresh : 0.001   #threshold on the condifence of the bbox
 verbose     : false #print on screen information
diff --git a/demo/demo/demo.cpp b/demo/demo/demo.cpp
index 76b451d..9f50d0b 100644
--- a/demo/demo/demo.cpp
+++ b/demo/demo/demo.cpp
@@ -40,6 +40,9 @@ int main(int argc, char *argv[]) {
     bool show = true;
     if(argc > 6)
         show = atoi(argv[6]); 
+    float conf_thresh=0.3;
+    if(argc > 7)
+        conf_thresh = atof(argv[7]);     
 
     if(n_batch < 1 || n_batch > 64)
         FatalError("Batch dim not supported");
@@ -69,7 +72,7 @@ int main(int argc, char *argv[]) {
         FatalError("Network type not allowed (3rd parameter)\n");
     }
 
-    detNN->init(net, n_classes, n_batch);
+    detNN->init(net, n_classes, n_batch, conf_thresh);
 
     gRun = true;
 
diff --git a/demo/demo/map.cpp b/demo/demo/map.cpp
index d724db0..356e35a 100644
--- a/demo/demo/map.cpp
+++ b/demo/demo/map.cpp
@@ -105,7 +105,7 @@ int main(int argc, char *argv[])
         default:
             FatalError("Network type not allowed (3rd parameter)\n");
     }
-    detNN->init(net, n_classes);
+    detNN->init(net, n_classes, 1, conf_thresh);
 
     //read images 
     std::ifstream all_labels(labels_path);
diff --git a/include/tkDNN/CenternetDetection.h b/include/tkDNN/CenternetDetection.h
index 227cb78..3c8cfbb 100644
--- a/include/tkDNN/CenternetDetection.h
+++ b/include/tkDNN/CenternetDetection.h
@@ -73,7 +73,7 @@ public:
     CenternetDetection() {};
     ~CenternetDetection() {}; 
 
-    bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1);
+    bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3);
     void preprocess(cv::Mat &frame, const int bi=0);
     void postprocess(const int bi=0,const bool mAP=false);
 };
diff --git a/include/tkDNN/DetectionNN.h b/include/tkDNN/DetectionNN.h
index 030cf8f..ba42834 100644
--- a/include/tkDNN/DetectionNN.h
+++ b/include/tkDNN/DetectionNN.h
@@ -84,7 +84,7 @@ class DetectionNN {
          * @param n_batches maximum number of batches to use in inference
          * @return true if everything is correct, false otherwise.
          */
-        virtual bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1) = 0;
+        virtual bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3) = 0;
         
         /**
          * This method performs the whole detection of the NN.
diff --git a/include/tkDNN/MobilenetDetection.h b/include/tkDNN/MobilenetDetection.h
index cabd7eb..9a5fedc 100644
--- a/include/tkDNN/MobilenetDetection.h
+++ b/include/tkDNN/MobilenetDetection.h
@@ -65,7 +65,7 @@ public:
     MobilenetDetection() {};
     ~MobilenetDetection() {}; 
 
-    bool init(const std::string& tensor_path, const int n_classes, const int n_batches=1);
+    bool init(const std::string& tensor_path, const int n_classes, const int n_batches=1, const float conf_thresh=0.3);
     void preprocess(cv::Mat &frame, const int bi=0);
     void postprocess(const int bi=0,const bool mAP=false);
 };
diff --git a/include/tkDNN/Yolo3Detection.h b/include/tkDNN/Yolo3Detection.h
index 6d38514..100a720 100644
--- a/include/tkDNN/Yolo3Detection.h
+++ b/include/tkDNN/Yolo3Detection.h
@@ -24,7 +24,7 @@ public:
     Yolo3Detection() {};
     ~Yolo3Detection() {}; 
 
-    bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1);
+    bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3);
     void preprocess(cv::Mat &frame, const int bi=0);
     void postprocess(const int bi=0,const bool mAP=false);
 };
diff --git a/src/CenternetDetection.cpp b/src/CenternetDetection.cpp
index 9d8df38..394e24a 100644
--- a/src/CenternetDetection.cpp
+++ b/src/CenternetDetection.cpp
@@ -3,11 +3,12 @@
 
 namespace tk { namespace dnn {
 
-bool CenternetDetection::init(const std::string& tensor_path, const int n_classes, const int n_batches){
+bool CenternetDetection::init(const std::string& tensor_path, const int n_classes, const int n_batches, const float conf_thresh){
     std::cout<<(tensor_path).c_str()<<"\n";
     netRT = new tk::dnn::NetworkRT(NULL, (tensor_path).c_str() );
     classes = n_classes;
     nBatches = n_batches;
+    confThreshold = conf_thresh;
 
     dim = netRT->input_dim;
 
diff --git a/src/MobilenetDetection.cpp b/src/MobilenetDetection.cpp
index c905fea..3c54e28 100644
--- a/src/MobilenetDetection.cpp
+++ b/src/MobilenetDetection.cpp
@@ -126,12 +126,13 @@ float MobilenetDetection::iou(const tk::dnn::box &a, const tk::dnn::box &b){
     return iou;
 }
 
-bool MobilenetDetection::init(const std::string& tensor_path, const int n_classes, const int n_batches){
+bool MobilenetDetection::init(const std::string& tensor_path, const int n_classes, const int n_batches, const float conf_thresh){
     std::cout<<(tensor_path).c_str()<<"\n";
     netRT = new tk::dnn::NetworkRT(NULL, (tensor_path).c_str());
     imageSize = netRT->input_dim.h;
     classes = n_classes;
     nBatches = n_batches;
+    confThreshold = conf_thresh;
 
     SSDSpec specs[N_SSDSPEC];
 
diff --git a/src/Yolo3Detection.cpp b/src/Yolo3Detection.cpp
index 606c6d1..e9b0064 100644
--- a/src/Yolo3Detection.cpp
+++ b/src/Yolo3Detection.cpp
@@ -3,13 +3,14 @@
 
 namespace tk { namespace dnn {
 
-bool Yolo3Detection::init(const std::string& tensor_path, const int n_classes, const int n_batches) {
+bool Yolo3Detection::init(const std::string& tensor_path, const int n_classes, const int n_batches, const float conf_thresh) {
 
     //convert network to tensorRT
     std::cout<<(tensor_path).c_str()<<"\n";
     netRT = new tk::dnn::NetworkRT(NULL, (tensor_path).c_str() );
 
     nBatches = n_batches;
+    confThreshold = conf_thresh;
     tk::dnn::dataDim_t idim = netRT->input_dim;    
     idim.n = nBatches;
 
@@ -109,10 +110,10 @@ void Yolo3Detection::postprocess(const int bi, const bool mAP){
     detected.clear();
     for(int j=0; j<nDets; j++) {
         tk::dnn::Yolo::box b = dets[j].bbox;
-        int x0   = (b.x-b.w/2.);
-        int x1   = (b.x+b.w/2.);
-        int y0   = (b.y-b.h/2.);
-        int y1   = (b.y+b.h/2.);
+        float x0   = (b.x-b.w/2.);
+        float x1   = (b.x+b.w/2.);
+        float y0   = (b.y-b.h/2.);
+        float y1   = (b.y+b.h/2.);
 
         // convert to image coords
         x0 = x_ratio*x0;

From 8bef544bae2e7f7d935abe0f7c986a78fd99218c Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Thu, 6 Aug 2020 11:05:26 +0200
Subject: [PATCH 14/53] Update README.md

---
 README.md | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/README.md b/README.md
index d9b5755..ea55785 100644
--- a/README.md
+++ b/README.md
@@ -14,7 +14,7 @@ M. Verucchi, G. Brilli, D. Sapienza, M. Verasani, M. Arena, F. Gatti, A. Capoton
 "A Systematic Assessment of Embedded Neural Networks for Object Detection", in IEEE International Conference on Emerging Technologies and Factory Automation (2020)
 ```
 
-## Results
+## FPS Results
 Inference FPS of yolov4 with tkDNN, average of 1200 images with the same dimesion as the input size, on 
   * RTX 2080Ti (CUDA 10.2, TensorRT 7.0.0, Cudnn 7.6.5);
   * Xavier AGX, Jetpack 4.3 (CUDA 10.0, CUDNN 7.6.3, tensorrt 6.0.1 );
@@ -40,6 +40,20 @@ Inference FPS of yolov4 with tkDNN, average of 1200 images with the same dimesio
 | Nano       | yolo4 512	| 2,32	    | 2,34	    | 3,02	    | 3,04      | -         | -         |
 | Nano       | yolo4 608	| 1,40	    | 1,41	    | 1,92	    | 1,93      | -         | -         |
 
+## MAP Results
+Results for COCO val 2017 (5k images), on RTX 2080Ti, with conf threshold=0.001
+
+|                      | CodaLab       | CodaLab   | CodaLab       | CodaLab     | tkDNN map     | tkDNN map |
+| -------------------- | :-----------: | :-------: | :-----------: | :---------: | :-----------: | :-------: |
+|                      | **tkDNN**     | **tkDNN** | **darknet**   | **darknet** | **tkDNN**     | **tkDNN** |
+|                      | MAP(0.5:0.95) | AP50      | MAP(0.5:0.95) | AP50        | MAP(0.5:0.95) | AP50      |
+| Yolov3 (416x416)     | 0.381         | 0.675     | 0.380         | 0.675       | 0.372         | 0.663     |
+| yolov4 (416x416)     | 0.468         | 0.705     | 0.471         | 0.710       | 0.459         | 0.695     |
+| yolov3tiny (416x416) | 0.096         | 0.202     | 0.096         | 0.201       | 0.093         | 0.198     |
+| yolov4tiny (416x416) | 0.202         | 0.400     | 0.201         | 0.400       | 0.197         | 0.395     |
+| Cnet-dla34 (512x512) | 0.366         | 0.543     | \-            | \-          | 0.361         | 0.535     |
+| mv2SSD (512x512)     | 0.226         | 0.381     | \-            | \-          | 0.223         | 0.378     |
+
 ## Index
 - [tkDNN](#tkdnn)
   - [Index](#index)

From df5443e017f9390b3f282a5c18ecf335ffafc5f8 Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Thu, 6 Aug 2020 15:53:57 +0200
Subject: [PATCH 15/53] Fix boxes also for Centernet

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 src/CenternetDetection.cpp | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/CenternetDetection.cpp b/src/CenternetDetection.cpp
index 394e24a..46757f4 100644
--- a/src/CenternetDetection.cpp
+++ b/src/CenternetDetection.cpp
@@ -372,10 +372,10 @@ void CenternetDetection::postprocess(const int bi, const bool mAP){
                     // std::cout<<"th: "<<scores[j]<<" - cl: "<<clses[j]<<" i: "<<i<<std::endl;
                     //add coco bbox
                     //det[0:4], i, det[4]
-                    int x0   = target_coords[j*4];
-                    int y0   = target_coords[j*4+1];
-                    int x1   = target_coords[j*4+2];
-                    int y1   = target_coords[j*4+3];
+                    float x0   = target_coords[j*4];
+                    float y0   = target_coords[j*4+1];
+                    float x1   = target_coords[j*4+2];
+                    float y1   = target_coords[j*4+3];
                     int obj_class = clses[j];
                     float prob = scores[j];
                     // std::cout<<"("<<x0<<", "<<y0<<"),("<<x1<<", "<<y1<<")"<<std::endl;

From fa2b3d26cb33395fa87549886c61c3f1c5dba62e Mon Sep 17 00:00:00 2001
From: micaela <micaelaverucchi@gmail.com>
Date: Fri, 11 Sep 2020 09:13:59 +0200
Subject: [PATCH 16/53] Fix typos (#107)

Signed-off-by: micaela <micaelaverucchi@gmail.com>
---
 README.md                                  | 14 +++++++-------
 include/tkDNN/DetectionNN.h                |  8 ++++----
 include/tkDNN/ImuOdom.h                    |  4 ++--
 include/tkDNN/Layer.h                      | 20 ++++++++++----------
 include/tkDNN/Network.h                    |  8 ++++----
 include/tkDNN/NetworkRT.h                  |  2 +-
 include/tkDNN/evaluation.h                 |  8 ++++----
 include/tkDNN/pluginsRT/DeformableConvRT.h |  2 +-
 include/tkDNN/test.h                       |  2 +-
 src/DarknetParser.cpp                      |  2 +-
 src/DeformConv2d.cpp                       |  2 +-
 src/Dense.cpp                              |  2 +-
 src/LSTM.cpp                               | 10 +++++-----
 src/LayerWgs.cpp                           |  2 +-
 src/MulAdd.cpp                             |  2 +-
 src/NetworkRT.cpp                          |  2 +-
 src/Region.cpp                             |  2 +-
 src/Shortcut.cpp                           |  2 +-
 src/evaluation.cpp                         |  6 +++---
 19 files changed, 50 insertions(+), 50 deletions(-)

diff --git a/README.md b/README.md
index ea55785..f17f75e 100644
--- a/README.md
+++ b/README.md
@@ -15,7 +15,7 @@ M. Verucchi, G. Brilli, D. Sapienza, M. Verasani, M. Arena, F. Gatti, A. Capoton
 ```
 
 ## FPS Results
-Inference FPS of yolov4 with tkDNN, average of 1200 images with the same dimesion as the input size, on 
+Inference FPS of yolov4 with tkDNN, average of 1200 images with the same dimension as the input size, on 
   * RTX 2080Ti (CUDA 10.2, TensorRT 7.0.0, Cudnn 7.6.5);
   * Xavier AGX, Jetpack 4.3 (CUDA 10.0, CUDNN 7.6.3, tensorrt 6.0.1 );
   * Tx2, Jetpack 4.2 (CUDA 10.0, CUDNN 7.3.1, tensorrt 5.0.6 );
@@ -169,7 +169,7 @@ tkDNN implement and easy parser for darknet cfg files, a network can be converte
 tk::dnn::Network *net = tk::dnn::darknetParser("yolov4.cfg", "yolov4/layers", "coco.names");
 net->print();
 ```
-All models from darknet are now parsed directly from cfg, you still need to export the weights with the descripted tools in the previus section.
+All models from darknet are now parsed directly from cfg, you still need to export the weights with the described tools in the previous section.
 <details>
   <summary>Supported layers</summary>
   convolutional
@@ -203,7 +203,7 @@ cmake .. -DDEBUG=True
 make
 ```
 
-Once you have succesfully created your rt file, run the demo: 
+Once you have successfully created your rt file, run the demo: 
 ```
 ./demo yolo4_fp32.rt ../demo/yolo_test.mp4 y
 ```
@@ -247,7 +247,7 @@ You should provide image_list.txt and label_list.txt, using training images. How
 ```
 bash scripts/download_validation.sh COCO
 ```
-to automatically download COCO2017 validation (inside demo folder) and create those needed file. Use BDD insted of COCO to download BDD validation. 
+to automatically download COCO2017 validation (inside demo folder) and create those needed file. Use BDD instead of COCO to download BDD validation. 
 
 Then a complete example using yolo3 and COCO dataset would be:
 ```
@@ -269,8 +269,8 @@ N.B.
 export TKDNN_BATCHSIZE=2
 # build tensorRT files
 ```
-This will create a TensorRT file with the desidered **max** batch size.
-The test will still run with a batch of 1, but the created tensorRT can manage the desidered batch size.
+This will create a TensorRT file with the desired **max** batch size.
+The test will still run with a batch of 1, but the created tensorRT can manage the desired batch size.
 
 ### Test batch Inference
 This will test the network with random input and check if the output of each batch is the same.
@@ -316,7 +316,7 @@ cd build
 ./map_demo dla34_cnet_FP32.rt c ../demo/COCO_val2017/all_labels.txt ../demo/config.yaml
 ```
 
-This demo also creates a json file named ```net_name_COCO_res.json``` containing all the detections computed. The detections are in COCO format, the correct format to subit the results to [CodaLab COCO detection challenge](https://competitions.codalab.org/competitions/20794#participate).
+This demo also creates a json file named ```net_name_COCO_res.json``` containing all the detections computed. The detections are in COCO format, the correct format to submit the results to [CodaLab COCO detection challenge](https://competitions.codalab.org/competitions/20794#participate).
 
 ## Existing tests and supported networks
 
diff --git a/include/tkDNN/DetectionNN.h b/include/tkDNN/DetectionNN.h
index ba42834..0498d41 100644
--- a/include/tkDNN/DetectionNN.h
+++ b/include/tkDNN/DetectionNN.h
@@ -76,10 +76,10 @@ class DetectionNN {
         ~DetectionNN(){};
 
         /**
-         * Method used to inialize the class, allocate memory and compute 
+         * Method used to initialize the class, allocate memory and compute 
          * needed data.
          * 
-         * @param tensor_path path to the rt file og the NN.
+         * @param tensor_path path to the rt file of the NN.
          * @param n_classes number of classes for the given dataset.
          * @param n_batches maximum number of batches to use in inference
          * @return true if everything is correct, false otherwise.
@@ -141,9 +141,9 @@ class DetectionNN {
         }      
 
         /**
-         * Method to draw boundixg boxes and labels on a frame.
+         * Method to draw bounding boxes and labels on a frame.
          * 
-         * @param frames orginal frame to draw bounding box on.
+         * @param frames original frame to draw bounding box on.
          */
         void draw(std::vector<cv::Mat>& frames) {
             tk::dnn::box b;
diff --git a/include/tkDNN/ImuOdom.h b/include/tkDNN/ImuOdom.h
index 58def96..6d8d4cb 100644
--- a/include/tkDNN/ImuOdom.h
+++ b/include/tkDNN/ImuOdom.h
@@ -44,7 +44,7 @@ class ImuOdom {
         virtual ~ImuOdom() {}
 
         /**
-         * Method used for inizialize the class
+         * Method used for initialize the class
          * 
          * @return Success of the initialization
          */
@@ -141,7 +141,7 @@ class ImuOdom {
             //odomPOS = odomPOS + deltaP.cast<double>(); // V2
             odomROT = odomROT * q.normalized().toRotationMatrix();
             
-            // compute euler
+            // compute Euler
             auto newEULER = odomROT.eulerAngles(0, 1, 2);
             for(int i=0; i<3; i++) {
                 while( fabs(newEULER(i) - odomEULER(i)) > M_PI_2 )  {
diff --git a/include/tkDNN/Layer.h b/include/tkDNN/Layer.h
index f2ec56d..790a431 100644
--- a/include/tkDNN/Layer.h
+++ b/include/tkDNN/Layer.h
@@ -171,7 +171,7 @@ public:
 
 
 /**
-    Input layer (it doesnt need weigths)
+    Input layer (it doesn't need weights)
 */
 class Input : public Layer {
 
@@ -207,7 +207,7 @@ public:
 
 
 /**
-    Avaible activation functions
+    Available activation functions
 */
 typedef enum {
     ACTIVATION_ELU     = 100,
@@ -216,7 +216,7 @@ typedef enum {
 } tkdnnActivationMode_t;
 
 /**
-    Activation layer (it doesnt need weigths)
+    Activation layer (it doesn't need weights)
 */
 class Activation : public Layer {
 
@@ -318,9 +318,9 @@ public:
     virtual dnnType* infer(dataDim_t &dim, dnnType* srcData);
 
     const bool bidirectional = true; /**> is the net bidir */
-    bool returnSeq = false;       /**> if false return only the result of last timestep */
+    bool returnSeq = false;       /**> if false return only the result of last timestamp */
     int stateSize = 0; /**> number of hidden states */
-    int seqLen = 0;    /**> number of timesteps */
+    int seqLen = 0;    /**> number of timestamp */
     int numLayers = 1; /**> number of internal layers */
 
 protected:
@@ -367,7 +367,7 @@ public:
 
 
 /**
-    Deformable Convolutionl 2d layer
+    Deformable Convolutional 2d layer
 */  
 class DeformConv2d : public LayerWgs {
 
@@ -449,7 +449,7 @@ protected:
 
 
 /**
-    Avaible pooling functions (padding on tkDNN is not supported)
+    Available pooling functions (padding on tkDNN is not supported)
 */
 typedef enum {
     POOLING_MAX     = 0,
@@ -460,7 +460,7 @@ typedef enum {
 
 /**
     Pooling layer
-    currenty supported only 2d pooing (also on 3d input)
+    currently supported only 2d pooing (also on 3d input)
 */
 class Pooling : public Layer {
 
@@ -526,7 +526,7 @@ public:
 
 /**
     Reorg layer
-    Mantain same dimension but change C*H*W distribution
+    Maintains same dimension but change C*H*W distribution
 */
 class Reorg : public Layer {
 
@@ -559,7 +559,7 @@ public:
 
 /**
     Upsample layer
-    Mantain same dimension but change C*H*W distribution
+    Maintains same dimension but change C*H*W distribution
 */
 class Upsample : public Layer {
 
diff --git a/include/tkDNN/Network.h b/include/tkDNN/Network.h
index 2d95215..b78acff 100644
--- a/include/tkDNN/Network.h
+++ b/include/tkDNN/Network.h
@@ -7,12 +7,12 @@
 namespace tk { namespace dnn {
 
 /**
-    Data rapresentation beetween layers
+    Data representation between layers
     n = batch size
     c = channels
-    h = heigth (lines)
+    h = height (lines)
     w = width  (rows)
-    l = lenght (3rd dimension)
+    l = length (3rd dimension)
 */
 struct dataDim_t {
 
@@ -43,7 +43,7 @@ public:
     void releaseLayers();
 
     /**
-        Do inferece for every added layer
+        Do inference for every added layer
     */
     dnnType* infer(dataDim_t &dim, dnnType* data);
 
diff --git a/include/tkDNN/NetworkRT.h b/include/tkDNN/NetworkRT.h
index 66b4f3d..4c6c816 100644
--- a/include/tkDNN/NetworkRT.h
+++ b/include/tkDNN/NetworkRT.h
@@ -91,7 +91,7 @@ public:
     }
 
     /**
-        Do inferece
+        Do inference
     */
     dnnType* infer(dataDim_t &dim, dnnType* data);
     void enqueue(int batchSize = 1);    
diff --git a/include/tkDNN/evaluation.h b/include/tkDNN/evaluation.h
index 8907d9d..128eba0 100644
--- a/include/tkDNN/evaluation.h
+++ b/include/tkDNN/evaluation.h
@@ -73,12 +73,12 @@ double computeMap(  std::vector<Frame> &images,const int classes,
  *                  all the recall levels are evaluated, otherwise only 
  *                  map_point recall levels are used. For COCO evaluation 
  *                  101 points are used.
- * @param map_step step used to increment IoU theshold
+ * @param map_step step used to increment IoU threshold
  * @param map_levels number of IoU step to perform
  * @param verbose is set to true, prints on screen additional info
  * @param write_on_file if set to true, the results produced by this function 
  *                      are written on file
- * @param net name of the considerd neural network 
+ * @param net name of the considered neural network 
  *
  * @return  mAP IoU_tresh:IoU_tresh+map_step*map_levels (e.g. mAP 0.5:0.95 when 
  *          map_step=0.05 and map_levels=10)
@@ -89,7 +89,7 @@ double computeMapNIoULevels(std::vector<Frame> &images,const int classes,
                 const int map_levels=10, const bool verbose=false, 
                 const bool write_on_file = false, std::string net = "");
 /**
- * This method computes the numper of True Positive (TP), False Positive (FP),
+ * This method computes the number of True Positive (TP), False Positive (FP),
  * False Negative (FN), precision, recall and f1-score.
  * Those values are computer over all the detections, over all the classes.
  *
@@ -101,7 +101,7 @@ double computeMapNIoULevels(std::vector<Frame> &images,const int classes,
  * @param verbose is set to true, prints on screen additional info
  * @param write_on_file if set to true, the results produced by this function 
  *                      are written on file
- * @param net name of the considerd neural network 
+ * @param net name of the considered neural network 
  */
 void computeTPFPFN( std::vector<Frame> &images,const int classes,
                     const float IoU_thresh=0.5, const float conf_thresh=0.3, 
diff --git a/include/tkDNN/pluginsRT/DeformableConvRT.h b/include/tkDNN/pluginsRT/DeformableConvRT.h
index bff6370..225a24e 100644
--- a/include/tkDNN/pluginsRT/DeformableConvRT.h
+++ b/include/tkDNN/pluginsRT/DeformableConvRT.h
@@ -89,7 +89,7 @@ public:
 		for(int b=0; b<batchSize; b++) {
 			checkCuda(cudaMemcpy(offset, output_conv + b * 3 * chunk_dim, 2*chunk_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice)); 
 			checkCuda(cudaMemcpy(mask, output_conv + b * 3 * chunk_dim + 2*chunk_dim, chunk_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice)); 
-			// kernel sigmoide
+			// kernel sigmoid
 			activationSIGMOIDForward(mask, mask, chunk_dim);
 			// deformable convolution
 			dcnV2CudaForward(stat, handle, 
diff --git a/include/tkDNN/test.h b/include/tkDNN/test.h
index 4e238ff..13c943e 100644
--- a/include/tkDNN/test.h
+++ b/include/tkDNN/test.h
@@ -20,7 +20,7 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
     }
     if(output_bins.size() != outputs.size()) {
         std::cout<<output_bins.size()<<" "<<outputs.size()<<"\n";
-        FatalError("outputs size missmatch");
+        FatalError("outputs size mismatch");
     }
 
     // Load input
diff --git a/src/DarknetParser.cpp b/src/DarknetParser.cpp
index 7bbc7ba..7d7d989 100644
--- a/src/DarknetParser.cpp
+++ b/src/DarknetParser.cpp
@@ -201,7 +201,7 @@ namespace tk { namespace dnn {
 
         tk::dnn::Network *net = nullptr;
         
-        // layers without activations to retrive correct id number
+        // layers without activations to retrieve correct id number
         std::vector<tk::dnn::Layer*> netLayers;
 
         std::ifstream if_cfg(cfg_file);
diff --git a/src/DeformConv2d.cpp b/src/DeformConv2d.cpp
index b161a22..dbb71e1 100644
--- a/src/DeformConv2d.cpp
+++ b/src/DeformConv2d.cpp
@@ -95,7 +95,7 @@ dnnType* DeformConv2d::infer(dataDim_t &dim, dnnType* srcData) {
     // split conv2d outputs into offset and mask
     checkCuda(cudaMemcpy(offset, output_conv, 2*chunk_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice)); 
     checkCuda(cudaMemcpy(mask, output_conv + 2*chunk_dim, chunk_dim*sizeof(dnnType), cudaMemcpyDeviceToDevice)); 
-    // kernel sigmoide
+    // kernel sigmoid
     activationSIGMOIDForward(mask, mask, chunk_dim);
 
     // deformable convolution
diff --git a/src/Dense.cpp b/src/Dense.cpp
index b6a9af2..4371d06 100644
--- a/src/Dense.cpp
+++ b/src/Dense.cpp
@@ -37,7 +37,7 @@ dnnType* Dense::infer(dataDim_t &dim, dnnType* srcData) {
     // place bias into dstData
     checkCuda( cudaMemcpy(dstData, bias_d, dim_y*sizeof(dnnType), cudaMemcpyDeviceToDevice) );
     
-    //do matrix moltiplication
+    //do matrix multiplication
     checkERROR( cublasSgemv(net->cublasHandle, CUBLAS_OP_T,
                             dim_x, dim_y,
                             &alpha,
diff --git a/src/LSTM.cpp b/src/LSTM.cpp
index 511fbee..7b87711 100644
--- a/src/LSTM.cpp
+++ b/src/LSTM.cpp
@@ -133,7 +133,7 @@ LSTM::LSTM( Network *net, int hiddensize, bool returnSeq, std::string fname_weig
     output_dim = input_dim;
     output_dim.c = stateSize*(bidirectional ? 2 : 1);
 
-    // if retunseq is disabled only the last timestep is returned
+    // if retunseq is disabled only the last timestamp is returned
     if(!returnSeq) {
         output_dim.h = 1;
         output_dim.w = 1;
@@ -254,7 +254,7 @@ dnnType* LSTM::infer(dataDim_t &dim, dnnType* srcData) {
             rnnDesc,
             seqLen,                     // number of time steps (nT)
             x_desc_vec_.data(),         // input array of desc (nT*nC_in)
-            srcF,                    // input pointer
+            srcF,                       // input pointer
             hx_desc_,                   // initial hidden state desc     
             hx_ptr,                     // initial hidden state pointer 
             cx_desc_,                   // initial cell state desc      
@@ -281,7 +281,7 @@ dnnType* LSTM::infer(dataDim_t &dim, dnnType* srcData) {
             rnnDesc,
             seqLen,                     // number of time steps (nT)
             x_desc_vec_.data(),         // input array of desc (nT*nC_in)
-            srcB,                    // input pointer
+            srcB,                       // input pointer
             hx_desc_,                   // initial hidden state desc     
             hx_ptr,                     // initial hidden state pointer 
             cx_desc_,                   // initial cell state desc      
@@ -289,7 +289,7 @@ dnnType* LSTM::infer(dataDim_t &dim, dnnType* srcData) {
             w_desc_,                    // weights desc
             wb_ptr,                     // weights pointer
             y_desc_vec_.data(),         // output desc     (nT*nC_out)
-            dstB_NR,                       // output pointer
+            dstB_NR,                    // output pointer
             hy_desc_,                   // final hidden state desc        
             hy_ptr,                     // final hidden state pointer 
             cy_desc_,                   // final cell state desc          
@@ -307,7 +307,7 @@ dnnType* LSTM::infer(dataDim_t &dim, dnnType* srcData) {
             one_output_dim.c*sizeof(dnnType), cudaMemcpyDeviceToDevice));
     }    
 
-    // if retunseq is disabled only the last timestep is returned
+    // if retunseq is disabled only the last timestamp is returned
     if(returnSeq) {
         // forward transpose
         matrixTranspose(net->cublasHandle, dstF, dstData, 
diff --git a/src/LayerWgs.cpp b/src/LayerWgs.cpp
index 4afb7cc..a761327 100644
--- a/src/LayerWgs.cpp
+++ b/src/LayerWgs.cpp
@@ -105,7 +105,7 @@ LayerWgs::LayerWgs(Network *net, int inputs, int outputs,
         float2half(tmp_d, variance16_d, b_size);
         cudaMemcpy(variance16_h, variance16_d, b_size*sizeof(__half), cudaMemcpyDeviceToHost);
 
-        //conver scales
+        //convert scales
         float2half(scales_d, scales16_d, b_size);
         cudaMemcpy(scales16_h, scales16_d, b_size*sizeof(__half), cudaMemcpyDeviceToHost);
     
diff --git a/src/MulAdd.cpp b/src/MulAdd.cpp
index 0c2a962..25cec8d 100644
--- a/src/MulAdd.cpp
+++ b/src/MulAdd.cpp
@@ -12,7 +12,7 @@ MulAdd::MulAdd(Network *net, dnnType mul, dnnType add) : Layer(net) {
 
     int size = input_dim.tot();
 
-    // create a vector with all value setted to add 
+    // create a vector with all value set to add 
     dnnType *add_vector_h = new dnnType[size];
     for(int i=0; i<size; i++)
         add_vector_h[i] = add;
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index a6725aa..9d38440 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -163,7 +163,7 @@ NetworkRT::NetworkRT(Network *net, const char *name) {
 	// note that indices are guaranteed to be less than IEngine::getNbBindings()
 	buf_input_idx = engineRT->getBindingIndex("data"); 
     buf_output_idx = engineRT->getBindingIndex("out");
-    std::cout<<"input idex = "<<buf_input_idx<<" -> output index = "<<buf_output_idx<<"\n";
+    std::cout<<"input index = "<<buf_input_idx<<" -> output index = "<<buf_output_idx<<"\n";
 
 
     Dims iDim = engineRT->getBindingDimensions(buf_input_idx);
diff --git a/src/Region.cpp b/src/Region.cpp
index 65bb786..7c26208 100644
--- a/src/Region.cpp
+++ b/src/Region.cpp
@@ -63,7 +63,7 @@ dnnType* Region::infer(dataDim_t &dim, dnnType* srcData) {
 }
 
 
-/* Intepret class */
+/* Interpret class */
 RegionInterpret::RegionInterpret(dataDim_t input_dim, dataDim_t output_dim, 
     int classes, int coords, int num, float thresh, std::string fname_weights) {
 
diff --git a/src/Shortcut.cpp b/src/Shortcut.cpp
index 78a2f23..2c7a4f4 100644
--- a/src/Shortcut.cpp
+++ b/src/Shortcut.cpp
@@ -13,7 +13,7 @@ Shortcut::Shortcut(Network *net, Layer *backLayer) : Layer(net) {
     if( /*backLayer->output_dim.c != input_dim.c ||*/
         backLayer->output_dim.w != input_dim.w ||
         backLayer->output_dim.h != input_dim.h   )
-        FatalError("Shortcut dim missmatch");
+        FatalError("Shortcut dim mismatch");
 }
 
 Shortcut::~Shortcut() {
diff --git a/src/evaluation.cpp b/src/evaluation.cpp
index 58c951d..f23c380 100644
--- a/src/evaluation.cpp
+++ b/src/evaluation.cpp
@@ -63,7 +63,7 @@ double computeMap(  std::vector<Frame> &images,const int classes,
     
     int gt_checked = 0;
 
-    // for each detection comput IoU with groundtruth and match detetcion and 
+    // for each detection compute IoU with groundtruth and match detetcion and 
     // groundtruth with IoU greater than IoU_thresh
     for(auto &img:images){
         for(size_t i=0; i<img.det.size(); i++){
@@ -153,7 +153,7 @@ double computeMap(  std::vector<Frame> &images,const int classes,
         }
     }
 
-    //compute average precision for each class. Two methods are avaible, 
+    //compute average precision for each class. Two methods are available, 
     //based on map_points required
     double mean_average_precision = 0;
     double last_recall, last_precision, delta_recall;
@@ -287,7 +287,7 @@ void computeTPFPFN( std::vector<Frame> &images,const int classes,
         }
     }
 
-    //count all TP, FP, FN and compute precsion, recall and f1-score
+    //count all TP, FP, FN and compute precision, recall and f1-score
     double avg_precision = 0, avg_recall = 0, f1_score = 0;
     int TP = 0, FP = 0, FN = 0;
     for(size_t i=0; i<classes; i++){

From be818d5e3ac8a30f8e6513ce09fee37d1519e63d Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Tue, 15 Sep 2020 10:50:24 +0200
Subject: [PATCH 17/53] Update README with Xavier NX FPS results

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 README.md | 37 +++++++++++++++++++++----------------
 1 file changed, 21 insertions(+), 16 deletions(-)

diff --git a/README.md b/README.md
index f17f75e..54b8140 100644
--- a/README.md
+++ b/README.md
@@ -18,27 +18,32 @@ M. Verucchi, G. Brilli, D. Sapienza, M. Verasani, M. Arena, F. Gatti, A. Capoton
 Inference FPS of yolov4 with tkDNN, average of 1200 images with the same dimension as the input size, on 
   * RTX 2080Ti (CUDA 10.2, TensorRT 7.0.0, Cudnn 7.6.5);
   * Xavier AGX, Jetpack 4.3 (CUDA 10.0, CUDNN 7.6.3, tensorrt 6.0.1 );
+  * Xavier NX, Jetpack 4.4  (CUDA 10.2, CUDNN 8.0.0, tensorrt 7.1.0 ). 
   * Tx2, Jetpack 4.2 (CUDA 10.0, CUDNN 7.3.1, tensorrt 5.0.6 );
   * Jetson Nano, Jetpack 4.4  (CUDA 10.2, CUDNN 8.0.0, tensorrt 7.1.0 ). 
 
 | Platform   | Network    | FP32, B=1 | FP32, B=4	| FP16, B=1 |	FP16, B=4 |	INT8, B=1 |	INT8, B=4 | 
 | :------:   | :-----:    | :-----:   | :-----:   | :-----:   |	:-----:   |	:-----:   |	:-----:   | 
-| RTX 2080Ti | yolo4  320 | 118,59	  |237,31	    | 207,81	  | 443,32	  | 262,37	  | 530,93    | 
-| RTX 2080Ti | yolo4  416 | 104,81	  |162,86	    | 169,06	  | 293,78	  | 206,93	  | 353,26    | 
-| RTX 2080Ti | yolo4  512 | 92,98	    |132,43	    | 140,36	  | 215,17	  | 165,35	  | 254,96    | 
-| RTX 2080Ti | yolo4  608 | 63,77	    |81,53	    | 111,39	  | 152,89	  | 127,79	  | 184,72    | 
-| AGX Xavier | yolo4 320  |	26,78	    |32,05	    | 57,14	    | 79,05	    | 73,15	    | 97,56     |
-| AGX Xavier | yolo4 416  |	19,96	    |21,52	    | 41,01	    | 49,00	    | 50,81	    | 60,61     |
-| AGX Xavier | yolo4 512  |	16,58	    |16,98	    | 31,12	    | 33,84	    | 37,82	    | 41,28     |
-| AGX Xavier | yolo4 608  |	9,45 	    |10,13	    | 21,92	    | 23,36	    | 27,05	    | 28,93     |
-| Tx2        | yolo4 320	| 11,18	    | 12,07	    | 15,32	    | 16,31     | -         | -         |
-| Tx2        | yolo4 416	| 7,30	    | 7,58	    | 9,45	    | 9,90      | -         | -         |
-| Tx2        | yolo4 512	| 5,96	    | 5,95	    | 7,22	    | 7,23      | -         | -         |
-| Tx2        | yolo4 608	| 3,63	    | 3,65	    | 4,67	    | 4,70      | -         | -         |
-| Nano       | yolo4 320	| 4,23	    | 4,55	    | 6,14	    | 6,53      | -         | -         |
-| Nano       | yolo4 416	| 2,88	    | 3,00	    | 3,90	    | 4,04      | -         | -         |
-| Nano       | yolo4 512	| 2,32	    | 2,34	    | 3,02	    | 3,04      | -         | -         |
-| Nano       | yolo4 608	| 1,40	    | 1,41	    | 1,92	    | 1,93      | -         | -         |
+| RTX 2080Ti | yolo4 320  | 118.59	  | 237.31	  | 207.81	  | 443.32	  | 262.37	  | 530.93    | 
+| RTX 2080Ti | yolo4 416  | 104.81	  | 162.86	  | 169.06	  | 293.78	  | 206.93	  | 353.26    | 
+| RTX 2080Ti | yolo4 512  | 92.98	    | 132.43	  | 140.36	  | 215.17	  | 165.35	  | 254.96    | 
+| RTX 2080Ti | yolo4 608  | 63.77	    | 81.53	    | 111.39	  | 152.89	  | 127.79	  | 184.72    | 
+| AGX Xavier | yolo4 320  |	26.78	    | 32.05	    | 57.14	    | 79.05	    | 73.15	    | 97.56     |
+| AGX Xavier | yolo4 416  |	19.96	    | 21.52	    | 41.01	    | 49.00	    | 50.81	    | 60.61     |
+| AGX Xavier | yolo4 512  |	16.58	    | 16.98	    | 31.12	    | 33.84	    | 37.82	    | 41.28     |
+| AGX Xavier | yolo4 608  |	9.45 	    | 10.13	    | 21.92	    | 23.36	    | 27.05	    | 28.93     |
+| Xavier NX  | yolo4 320  |	11.49	    | 13.79     |	25.26    	| 35.51	    | 33.77	    | 45.66     |
+| Xavier NX  | yolo4 416  |	8.38	    | 9.65	    | 18.72	    | 22.97	    | 23.71	    | 29.28     |
+| Xavier NX  | yolo4 512  |	7.03	    | 7.57	    | 13.50	    | 15.43	    | 17.95	    | 19.70     |
+| Xavier NX  | yolo4 608  |	4.49	    | 4.56	    | 10.10	    | 11.02	    | 13.09	    | 14.04     |
+| Tx2        | yolo4 320	| 11.18	    | 12.07	    | 15.32	    | 16.31     | -         | -         |
+| Tx2        | yolo4 416	| 7.30	    | 7.58	    | 9.45	    | 9.90      | -         | -         |
+| Tx2        | yolo4 512	| 5.96	    | 5.95	    | 7.22	    | 7.23      | -         | -         |
+| Tx2        | yolo4 608	| 3.63	    | 3.65	    | 4.67	    | 4.70      | -         | -         |
+| Nano       | yolo4 320	| 4.23	    | 4.55	    | 6.14	    | 6.53      | -         | -         |
+| Nano       | yolo4 416	| 2.88	    | 3.00	    | 3.90	    | 4.04      | -         | -         |
+| Nano       | yolo4 512	| 2.32	    | 2.34	    | 3.02	    | 3.04      | -         | -         |
+| Nano       | yolo4 608	| 1.40	    | 1.41	    | 1.92	    | 1.93      | -         | -         |
 
 ## MAP Results
 Results for COCO val 2017 (5k images), on RTX 2080Ti, with conf threshold=0.001

From d3372aad31d27d68593209f13e7c189752ec6e42 Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Tue, 15 Sep 2020 14:09:02 +0200
Subject: [PATCH 18/53] Update README with Xavier NX FPS results 15W4Core

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 README.md | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/README.md b/README.md
index 54b8140..70ddd26 100644
--- a/README.md
+++ b/README.md
@@ -32,10 +32,10 @@ Inference FPS of yolov4 with tkDNN, average of 1200 images with the same dimensi
 | AGX Xavier | yolo4 416  |	19.96	    | 21.52	    | 41.01	    | 49.00	    | 50.81	    | 60.61     |
 | AGX Xavier | yolo4 512  |	16.58	    | 16.98	    | 31.12	    | 33.84	    | 37.82	    | 41.28     |
 | AGX Xavier | yolo4 608  |	9.45 	    | 10.13	    | 21.92	    | 23.36	    | 27.05	    | 28.93     |
-| Xavier NX  | yolo4 320  |	11.49	    | 13.79     |	25.26    	| 35.51	    | 33.77	    | 45.66     |
-| Xavier NX  | yolo4 416  |	8.38	    | 9.65	    | 18.72	    | 22.97	    | 23.71	    | 29.28     |
-| Xavier NX  | yolo4 512  |	7.03	    | 7.57	    | 13.50	    | 15.43	    | 17.95	    | 19.70     |
-| Xavier NX  | yolo4 608  |	4.49	    | 4.56	    | 10.10	    | 11.02	    | 13.09	    | 14.04     |
+| Xavier NX  | yolo4 320  |	14.56	    | 16.25	    | 30.14	    | 41.15	    | 42.13	    | 53.42     |
+| Xavier NX  | yolo4 416  |	10.02	    | 10.60	    | 22.43	    | 25.59	    | 29.08	    | 32.94     |
+| Xavier NX  | yolo4 512  |	8.10	    | 8.32	    | 15.78	    | 17.13	    | 20.51	    | 22.46     |
+| Xavier NX  | yolo4 608  |	5.26	    | 5.18	    | 11.54	    | 12.06	    | 15.09	    | 15.82     |
 | Tx2        | yolo4 320	| 11.18	    | 12.07	    | 15.32	    | 16.31     | -         | -         |
 | Tx2        | yolo4 416	| 7.30	    | 7.58	    | 9.45	    | 9.90      | -         | -         |
 | Tx2        | yolo4 512	| 5.96	    | 5.95	    | 7.22	    | 7.23      | -         | -         |

From a0e7f05a50e5bc639a3c843139c39884d2c5a7fc Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Sat, 10 Oct 2020 13:03:01 +0200
Subject: [PATCH 19/53] Update README.md

---
 README.md | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)

diff --git a/README.md b/README.md
index 70ddd26..cdfa25b 100644
--- a/README.md
+++ b/README.md
@@ -3,15 +3,18 @@ tkDNN is a Deep Neural Network library built with cuDNN and tensorRT primitives,
 The main goal of this project is to exploit NVIDIA boards as much as possible to obtain the best inference performance. It does not allow training. 
 
 
-If you use tkDNN in your research, please cite one of the following papers. For use in commercial solutions, write at gattifrancesco@hotmail.it and micaela.verucchi@unimore.it or refer to https://hipert.unimore.it/ .
+If you use tkDNN in your research, please cite the [following paper](https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9212130&casa_token=sQTJXi7tJNoAAAAA:BguH9xCIY48MxbtDS3LXzIXzO-9sWArm7Hd7y7BwaLmqRuM_Gx8bOYizFPNMNtpo5K0kB-P-). For use in commercial solutions, write at gattifrancesco@hotmail.it and micaela.verucchi@unimore.it or refer to https://hipert.unimore.it/ .
 
 ```
-Accepted paper @ IRC 2020, will soon be published.
-M. Verucchi, L. Bartoli, F. Bagni, F. Gatti, P. Burgio and M. Bertogna, "Real-Time clustering and LiDAR-camera fusion on embedded platforms for self-driving cars",  in proceedings in IEEE Robotic Computing (2020)
-
-Accepted paper @ ETFA 2020, will soon be published.
-M. Verucchi, G. Brilli, D. Sapienza, M. Verasani, M. Arena, F. Gatti, A. Capotondi, R. Cavicchioli, M. Bertogna, M. Solieri
-"A Systematic Assessment of Embedded Neural Networks for Object Detection", in IEEE International Conference on Emerging Technologies and Factory Automation (2020)
+@inproceedings{verucchi2020systematic,
+  title={A Systematic Assessment of Embedded Neural Networks for Object Detection},
+  author={Verucchi, Micaela and Brilli, Gianluca and Sapienza, Davide and Verasani, Mattia and Arena, Marco and Gatti, Francesco and Capotondi, Alessandro and Cavicchioli, Roberto and Bertogna, Marko and Solieri, Marco},
+  booktitle={2020 25th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA)},
+  volume={1},
+  pages={937--944},
+  year={2020},
+  organization={IEEE}
+}
 ```
 
 ## FPS Results

From 86478f9384eef13d68a9406ee12fbcb4df6ab892 Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Fri, 23 Oct 2020 11:40:55 +0200
Subject: [PATCH 20/53] Add yolo4_mmr test

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 tests/darknet/cfg/yolo4_mmr.cfg | 1158 +++++++++++++++++++++++++++++++
 tests/darknet/names/mmr.names   |    4 +
 tests/darknet/yolo4_mmr.cpp     |   34 +
 3 files changed, 1196 insertions(+)
 create mode 100644 tests/darknet/cfg/yolo4_mmr.cfg
 create mode 100644 tests/darknet/names/mmr.names
 create mode 100644 tests/darknet/yolo4_mmr.cpp

diff --git a/tests/darknet/cfg/yolo4_mmr.cfg b/tests/darknet/cfg/yolo4_mmr.cfg
new file mode 100644
index 0000000..90a7204
--- /dev/null
+++ b/tests/darknet/cfg/yolo4_mmr.cfg
@@ -0,0 +1,1158 @@
+[net]
+batch=1
+subdivisions=1
+# Training
+width=512
+height=512
+# width=608
+# height=608
+channels=3
+momentum=0.949
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+learning_rate=0.0013
+burn_in=1000
+max_batches = 16000
+policy=steps
+steps=12800,14400
+scales=.1,.1
+
+#cutmix=1
+mosaic=1
+
+#:104x104 54:52x52 85:26x26 104:13x13 for 416
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-7
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-10
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-28
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-28
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-16
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=mish
+
+##########################
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+### SPP ###
+[maxpool]
+stride=1
+size=5
+
+[route]
+layers=-2
+
+[maxpool]
+stride=1
+size=9
+
+[route]
+layers=-4
+
+[maxpool]
+stride=1
+size=13
+
+[route]
+layers=-1,-3,-5,-6
+### End SPP ###
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[upsample]
+stride=2
+
+[route]
+layers = 85
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1, -3
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[upsample]
+stride=2
+
+[route]
+layers = 54
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[route]
+layers = -1, -3
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=leaky
+
+##########################
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=27
+activation=linear
+
+
+[yolo]
+mask = 0,1,2
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=4
+num=9
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+scale_x_y = 1.2
+iou_thresh=0.213
+cls_normalizer=1.0
+iou_normalizer=0.07
+iou_loss=ciou
+nms_kind=greedynms
+beta_nms=0.6
+max_delta=5
+
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=2
+pad=1
+filters=256
+activation=leaky
+
+[route]
+layers = -1, -16
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=27
+activation=linear
+
+
+[yolo]
+mask = 3,4,5
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=4
+num=9
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+scale_x_y = 1.1
+iou_thresh=0.213
+cls_normalizer=1.0
+iou_normalizer=0.07
+iou_loss=ciou
+nms_kind=greedynms
+beta_nms=0.6
+max_delta=5
+
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=2
+pad=1
+filters=512
+activation=leaky
+
+[route]
+layers = -1, -37
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=leaky
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=27
+activation=linear
+
+
+[yolo]
+mask = 6,7,8
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=4
+num=9
+jitter=.3
+ignore_thresh = .7
+truth_thresh = 1
+random=1
+scale_x_y = 1.05
+iou_thresh=0.213
+cls_normalizer=1.0
+iou_normalizer=0.07
+iou_loss=ciou
+nms_kind=greedynms
+beta_nms=0.6
+max_delta=5
+
diff --git a/tests/darknet/names/mmr.names b/tests/darknet/names/mmr.names
new file mode 100644
index 0000000..701a1fc
--- /dev/null
+++ b/tests/darknet/names/mmr.names
@@ -0,0 +1,4 @@
+blue-cone
+yellow-cone
+orange-cone
+big-orange-cone
\ No newline at end of file
diff --git a/tests/darknet/yolo4_mmr.cpp b/tests/darknet/yolo4_mmr.cpp
new file mode 100644
index 0000000..85649b2
--- /dev/null
+++ b/tests/darknet/yolo4_mmr.cpp
@@ -0,0 +1,34 @@
+#include<iostream>
+#include<vector>
+#include "tkdnn.h"
+#include "test.h"
+#include "DarknetParser.h"
+
+int main() {
+    std::string bin_path  = "yolo4_mmr";
+    std::vector<std::string> input_bins = { 
+        bin_path + "/layers/input.bin"
+    };
+    std::vector<std::string> output_bins = {
+        bin_path + "/debug/layer139_out.bin",
+        bin_path + "/debug/layer150_out.bin",
+        bin_path + "/debug/layer161_out.bin"
+    };
+    std::string wgs_path  = bin_path + "/layers";
+    std::string cfg_path  = std::string(TKDNN_PATH) + "/tests/darknet/cfg/yolo4_mmr.cfg";
+    std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/mmr.names";
+    // downloadWeightsifDoNotExist(input_bins[0], bin_path, "");
+
+    // parse darknet network
+    tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
+    net->print();
+
+    //convert network to tensorRT
+    tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
+    
+    int ret = testInference(input_bins, output_bins, net, netRT);
+    net->releaseLayers();
+    delete net;
+    delete netRT;
+    return ret;
+}

From 702791e41ac302ed0396034cf80d6d76303ac20a Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Mon, 23 Nov 2020 11:25:52 +0100
Subject: [PATCH 21/53] Add support for yolov4x-mish. Changes: - add parameters
 nms_kind, nms_thresh, new_coords to yolo layer and darknet parser - added
 diou nms, new method to compute the BBs - created test for yolov4x-mish
 called yolo4x

Tested, all tests work.
Problem to solve: little loss in mAP of yolo4x

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 include/tkDNN/DarknetParser.h    |    3 +
 include/tkDNN/Layer.h            |   14 +-
 include/tkDNN/pluginsRT/YoloRT.h |   20 +-
 scripts/test_all_tests.sh        |    1 +
 src/DarknetParser.cpp            |   14 +-
 src/NetworkRT.cpp                |   12 +-
 src/Yolo.cpp                     |   68 +-
 src/Yolo3Detection.cpp           |    7 +-
 tests/darknet/cfg/yolo4x.cfg     | 1427 ++++++++++++++++++++++++++++++
 tests/darknet/yolo4x.cpp         |   36 +
 10 files changed, 1571 insertions(+), 31 deletions(-)
 create mode 100644 tests/darknet/cfg/yolo4x.cfg
 create mode 100644 tests/darknet/yolo4x.cpp

diff --git a/include/tkDNN/DarknetParser.h b/include/tkDNN/DarknetParser.h
index 29d1e8e..089c4d6 100644
--- a/include/tkDNN/DarknetParser.h
+++ b/include/tkDNN/DarknetParser.h
@@ -24,7 +24,10 @@ namespace tk { namespace dnn {
         int num = 1;
         int pad = 0;
         int coords = 4;
+        int nms_kind = 0;
+        int new_coords= 0;
         float scale_xy = 1;
+        float nms_thresh = 0.45;
         std::vector<int> layers;
         std::string activation = "linear";
 
diff --git a/include/tkDNN/Layer.h b/include/tkDNN/Layer.h
index 790a431..25c4565 100644
--- a/include/tkDNN/Layer.h
+++ b/include/tkDNN/Layer.h
@@ -610,24 +610,28 @@ public:
         int sort_class;
     };
 
-    Yolo(Network *net, int classes, int num, std::string fname_weights,int n_masks=3, float scale_xy=1);
+    enum nmsKind_t {GREEDY_NMS=0, DIOU_NMS=1};
+
+    Yolo(Network *net, int classes, int num, std::string fname_weights,int n_masks=3, float scale_xy=1, double nms_thresh=0.45, nmsKind_t nsm_kind=GREEDY_NMS, int new_coords=0);
     virtual ~Yolo();
     virtual layerType_t getLayerType() { return LAYER_YOLO; };
 
-    int classes, num, n_masks;
+    int classes, num, n_masks, new_coords;
     dnnType *mask_h, *mask_d; //anchors
     dnnType *bias_h, *bias_d; //anchors
     float scaleXY;
+    double nms_thresh;
+    nmsKind_t nsm_kind; 
     std::vector<std::string> classesNames;
 
     virtual dnnType* infer(dataDim_t &dim, dnnType* srcData);
-    int computeDetections(Yolo::detection *dets, int &ndets, int netw, int neth, float thresh);
+    int computeDetections(Yolo::detection *dets, int &ndets, int netw, int neth, float thresh, int new_coords=0);
 
     dnnType *predictions;
 
-    static const int MAX_DETECTIONS = 8192;
+    static const int MAX_DETECTIONS = 8192*2;
     static Yolo::detection *allocateDetections(int nboxes, int classes);
-    static void             mergeDetections(Yolo::detection *dets, int ndets, int classes);
+    static void             mergeDetections(Yolo::detection *dets, int ndets, int classes, double nms_thresh=0.45, nmsKind_t nsm_kind=GREEDY_NMS);
 };
 
 /**
diff --git a/include/tkDNN/pluginsRT/YoloRT.h b/include/tkDNN/pluginsRT/YoloRT.h
index f8e596c..9af8587 100644
--- a/include/tkDNN/pluginsRT/YoloRT.h
+++ b/include/tkDNN/pluginsRT/YoloRT.h
@@ -8,12 +8,15 @@ class YoloRT : public IPlugin {
 
 
 public:
-	YoloRT(int classes, int num, tk::dnn::Yolo *yolo = nullptr, int n_masks=3, float scale_xy=1) {
+	YoloRT(int classes, int num, tk::dnn::Yolo *yolo = nullptr, int n_masks=3, float scale_xy=1, float nms_thresh=0.45, int nms_kind=0, int new_coords=0) {
 
 		this->classes = classes;
 		this->num = num;
 		this->n_masks = n_masks;
 		this->scaleXY = scale_xy;
+		this->nms_thresh = nms_thresh;
+		this->nms_kind = nms_kind;
+		this->new_coords = new_coords;
 
         mask = new dnnType[n_masks];
         bias = new dnnType[num*n_masks*2];
@@ -64,7 +67,10 @@ public:
 		for (int b = 0; b < batchSize; ++b){
 			for(int n = 0; n < n_masks; ++n){
 				int index = entry_index(b, n*w*h, 0);
-				activationLOGISTICForward(srcData + index, dstData + index, 2*w*h, stream);
+				if (new_coords == 1)
+					activationLOGISTICForward(srcData + index, dstData + index, 4*w*h, stream); //x,y,w,h
+				else
+					activationLOGISTICForward(srcData + index, dstData + index, 2*w*h, stream); //x,y
 
 				if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
 				
@@ -79,7 +85,7 @@ public:
 
 
 	virtual size_t getSerializationSize() override {
-		return 6*sizeof(int) + sizeof(float)+ n_masks*sizeof(dnnType) + num*n_masks*2*sizeof(dnnType) + YOLORT_CLASSNAME_W*classes*sizeof(char);
+		return 8*sizeof(int) + 2*sizeof(float)+ n_masks*sizeof(dnnType) + num*n_masks*2*sizeof(dnnType) + YOLORT_CLASSNAME_W*classes*sizeof(char);
 	}
 
 	virtual void serialize(void* buffer) override {
@@ -87,10 +93,13 @@ public:
 		tk::dnn::writeBUF(buf, classes);
 		tk::dnn::writeBUF(buf, num);
 		tk::dnn::writeBUF(buf, n_masks);
+		tk::dnn::writeBUF(buf, scaleXY);
+		tk::dnn::writeBUF(buf, nms_thresh);
+		tk::dnn::writeBUF(buf, nms_kind);
+		tk::dnn::writeBUF(buf, new_coords);
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
-		tk::dnn::writeBUF(buf, scaleXY);
         for(int i=0; i<n_masks; i++)
     		tk::dnn::writeBUF(buf, mask[i]);
         for(int i=0; i<n_masks*2*num; i++)
@@ -109,6 +118,9 @@ public:
 	int c, h, w;
     int classes, num, n_masks;
 	float scaleXY;
+	float nms_thresh;
+	int nms_kind;
+	int new_coords;
 	std::vector<std::string> classesNames;
 
     dnnType *mask;
diff --git a/scripts/test_all_tests.sh b/scripts/test_all_tests.sh
index 770aa22..af04aff 100644
--- a/scripts/test_all_tests.sh
+++ b/scripts/test_all_tests.sh
@@ -73,6 +73,7 @@ do
     print_output $? imuodom
 
     test_net yolo4
+    test_net yolo4x
     test_net yolo4_berkeley
     test_net yolo4tiny
     test_net yolo3
diff --git a/src/DarknetParser.cpp b/src/DarknetParser.cpp
index 7d7d989..7b5410c 100644
--- a/src/DarknetParser.cpp
+++ b/src/DarknetParser.cpp
@@ -37,7 +37,10 @@ namespace tk { namespace dnn {
         std::string name,value;
         if(!divideNameAndValue(line, name, value))
             return false;
-        if(name.find("width") !=  std::string::npos)
+
+        if(name.find("new_coords") !=  std::string::npos)
+            fields.new_coords = std::stoi(value);
+        else if(name.find("width") !=  std::string::npos)
             fields.width = std::stoi(value);
         else if(name.find("height") !=  std::string::npos)
             fields.height = std::stoi(value);
@@ -79,6 +82,13 @@ namespace tk { namespace dnn {
             fields.group_id = std::stoi(value);
         else if(name.find("scale_x_y") !=  std::string::npos)
             fields.scale_xy = std::stof(value);
+        else if(name.find("beta_nms") !=  std::string::npos)
+            fields.nms_thresh = std::stof(value);
+        else if(name.find("nms_kind") !=  std::string::npos){
+            if(value == "greedynms") fields.nms_kind = 0;
+            else if(value == "diounms") fields.nms_kind = 1;
+            else std::cout<<"Not supported nms_kind "<<value<<", setting to greedynms"<<std::endl;
+        }
         else if(name.find("from") !=  std::string::npos)
             fields.layers.push_back(std::stof(value));
         else if(name.find("mask") !=  std::string::npos){
@@ -161,7 +171,7 @@ namespace tk { namespace dnn {
         } else if(f.type == "yolo") {
             std::string wgs = wgs_path + "/g" + std::to_string(netLayers.size()) + ".bin";
             //printf("%d %d %s %d %f\n", f.classes, f.num/f.n_mask, wgs.c_str(), f.n_mask, f.scale_xy);
-            tk::dnn::Yolo *l = new tk::dnn::Yolo(net, f.classes, f.num/f.n_mask, wgs, f.n_mask, f.scale_xy);
+            tk::dnn::Yolo *l = new tk::dnn::Yolo(net, f.classes, f.num/f.n_mask, wgs, f.n_mask, f.scale_xy, f.nms_thresh, (tk::dnn::Yolo::nmsKind_t) f.nms_kind, f.new_coords);
             if(names.size() != f.classes)
                 FatalError("Mismatch between number of classes and names");
             l->classesNames = names;
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index 9d38440..501ade4 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -529,7 +529,7 @@ ILayer* NetworkRT::convert_layer(ITensor *input, Yolo *l) {
     //std::cout<<"convert Yolo\n";
 
     //std::cout<<"New plugin YOLO\n";
-    IPlugin *plugin = new YoloRT(l->classes, l->num, l, l->n_masks, l->scaleXY);
+    IPlugin *plugin = new YoloRT(l->classes, l->num, l, l->n_masks, l->scaleXY, l->nms_thresh, l->nsm_kind, l->new_coords);
     IPluginLayer *lRT = networkRT->addPlugin(&input, 1, *plugin);
     checkNULL(lRT);
     return lRT;
@@ -739,12 +739,16 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
     if(name.find("Yolo") == 0) {
         YoloRT *r = new YoloRT(readBUF<int>(buf),    //classes
                                 readBUF<int>(buf),   //num
-                                nullptr,
-                                readBUF<int>(buf));   //n_masks
+                                nullptr, //yolo
+                                readBUF<int>(buf), //n_masks
+                                readBUF<float>(buf), //scale_xy
+                                readBUF<float>(buf),  //nms_thresh 
+                                readBUF<int>(buf),  //nms_kind
+                                readBUF<int>(buf)  //new_coords
+                                );   
         r->c = readBUF<int>(buf);
         r->h = readBUF<int>(buf);
         r->w = readBUF<int>(buf);
-        r->scaleXY = readBUF<float>(buf);
         for(int i=0; i<r->n_masks; i++)
             r->mask[i] = readBUF<dnnType>(buf);
         for(int i=0; i<r->n_masks*2*r->num; i++)
diff --git a/src/Yolo.cpp b/src/Yolo.cpp
index a4416be..9737e74 100644
--- a/src/Yolo.cpp
+++ b/src/Yolo.cpp
@@ -11,7 +11,7 @@
 
 namespace tk { namespace dnn {
 
-Yolo::Yolo(Network *net, int classes, int num, std::string fname_weights, int n_masks, float scale_xy) : 
+Yolo::Yolo(Network *net, int classes, int num, std::string fname_weights, int n_masks, float scale_xy, double nms_thresh, nmsKind_t nsm_kind, int new_coords) : 
     Layer(net) {
     this->final = true;
 
@@ -19,6 +19,9 @@ Yolo::Yolo(Network *net, int classes, int num, std::string fname_weights, int n_
     this->num = num;
     this->n_masks = n_masks;
     this->scaleXY = scale_xy;
+    this->nms_thresh = nms_thresh;
+    this->nsm_kind = nsm_kind;
+    this->new_coords = new_coords;
 
     // load anchors
     if(fname_weights != "") {
@@ -59,12 +62,21 @@ int entry_index(int batch, int location, int entry,
            entry*input_dim.w*input_dim.h + loc;
 }
 
-Yolo::box get_yolo_box(float *x, float *biases, int n, int index, int i, int j, int lw, int lh, int w, int h, int stride) {
+Yolo::box get_yolo_box(float *x, float *biases, int n, int index, int i, int j, int lw, int lh, int w, int h, int stride, int new_coords) {
     Yolo::box b;
-    b.x = (i + x[index + 0*stride]) / lw;
-    b.y = (j + x[index + 1*stride]) / lh;
-    b.w = exp(x[index + 2*stride]) * biases[2*n]   / w;
-    b.h = exp(x[index + 3*stride]) * biases[2*n+1] / h;
+
+    if(new_coords == 0){
+        b.x = (i + x[index + 0*stride]) / lw;
+        b.y = (j + x[index + 1*stride]) / lh;
+        b.w = exp(x[index + 2*stride]) * biases[2*n]   / w;
+        b.h = exp(x[index + 3*stride]) * biases[2*n+1] / h;
+    }
+    else{
+        b.x = (i + x[index + 0 * stride] * 2 - 0.5) / lw;
+        b.y = (j + x[index + 1 * stride] * 2 - 0.5) / lh;
+        b.w = x[index + 2 * stride] * x[index + 2 * stride] * 4 * biases[2 * n] / w;
+        b.h = x[index + 3 * stride] * x[index + 3 * stride] * 4 * biases[2 * n + 1] / h;
+    }
     return b;
 }
 
@@ -75,7 +87,10 @@ dnnType* Yolo::infer(dataDim_t &dim, dnnType* srcData) {
     for (int b = 0; b < dim.n; ++b){
         for(int n = 0; n < n_masks; ++n){
             int index = entry_index(b, n*dim.w*dim.h, 0, classes, input_dim, output_dim);
-            activationLOGISTICForward(srcData + index, dstData + index, 2*dim.w*dim.h);
+            if (new_coords == 1)
+                activationLOGISTICForward(srcData + index, dstData + index, 4*dim.w*dim.h);
+            else
+                activationLOGISTICForward(srcData + index, dstData + index, 2*dim.w*dim.h);
 
             if (this->scaleXY != 1) scalAdd(dstData + index, 2 * dim.w*dim.h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
             
@@ -116,7 +131,7 @@ void correct_yolo_boxes(Yolo::detection *dets, int n, int w, int h, int netw, in
     }
 }
 
-int Yolo::computeDetections(Yolo::detection *dets, int &ndets, int netw, int neth, float thresh) {
+int Yolo::computeDetections(Yolo::detection *dets, int &ndets, int netw, int neth, float thresh, int new_coords) {
 
     if(predictions == nullptr)
         predictions = new dnnType[output_dim.tot()];
@@ -140,7 +155,7 @@ int Yolo::computeDetections(Yolo::detection *dets, int &ndets, int netw, int net
             if(objectness <= thresh) continue;
             int box_index  = entry_index(0, n*lw*lh + i, 0, classes, input_dim, output_dim);
             
-            dets[count].bbox = get_yolo_box(predictions, bias_h, mask_h[n], box_index, col, row, lw, lh, netw, neth, lw*lh);
+            dets[count].bbox = get_yolo_box(predictions, bias_h, mask_h[n], box_index, col, row, lw, lh, netw, neth, lw*lh, new_coords);
             dets[count].objectness = objectness;
             dets[count].classes = classes;
             for(j = 0; j < classes; ++j){
@@ -193,6 +208,32 @@ float yolo_box_iou(Yolo::box a, Yolo::box b)
     return yolo_box_intersection(a, b)/yolo_box_union(a, b);
 }
 
+void box_c(const Yolo::box a, const Yolo::box b, float& top, float& bot, float& left, float& right) {
+    top = std::min(a.y - a.h / 2, b.y - b.h / 2);
+    bot = std::max(a.y + a.h / 2, b.y + b.h / 2);
+    left = std::min(a.x - a.w / 2, b.x - b.w / 2);
+    right = std::max(a.x + a.w / 2, b.x + b.w / 2);
+}
+
+// https://github.com/Zzh-tju/DIoU-darknet
+// https://arxiv.org/abs/1911.08287
+float yolo_box_diou(const Yolo::box a, const Yolo::box b, const float nms_thresh=0.6)
+{
+    float top, bot, left, right;
+    box_c(a, b, top, bot, left, right);
+    float w = right - left;
+    float h = bot - top;
+    float c = w * w + h * h;
+    float iou = yolo_box_iou(a, b);
+    if (c == 0) 
+        return iou;
+    
+    float d = (a.x - b.x) * (a.x - b.x) + (a.y - b.y) * (a.y - b.y);
+    float u = pow(d / c, nms_thresh);
+    float diou_term = u;
+    return iou - diou_term;
+}
+
 int yolo_nms_comparator(const void *pa, const void *pb)
 {
     Yolo::detection a = *(Yolo::detection *)pa;
@@ -219,8 +260,7 @@ Yolo::detection *Yolo::allocateDetections(int nboxes, int classes) {
     return dets;
 }
 
-void Yolo::mergeDetections(Yolo::detection *dets, int ndets, int classes) {
-    double nms_thresh = 0.45;
+void Yolo::mergeDetections(Yolo::detection *dets, int ndets, int classes, double nms_thresh, nmsKind_t nsm_kind) {
     int total = ndets;
 
     int i, j, k;
@@ -246,13 +286,13 @@ void Yolo::mergeDetections(Yolo::detection *dets, int ndets, int classes) {
             box a = dets[i].bbox;
             for(j = i+1; j < total; ++j){
                 box b = dets[j].bbox;
-                if (yolo_box_iou(a, b) > nms_thresh){
+                if (nsm_kind == GREEDY_NMS && yolo_box_iou(a, b) > nms_thresh)
+                    dets[j].prob[k] = 0;
+                else if (nsm_kind == DIOU_NMS && yolo_box_diou(a, b, nms_thresh) > nms_thresh)
                     dets[j].prob[k] = 0;
-                }
             }
         }
     }
-
 }
 
 }}
diff --git a/src/Yolo3Detection.cpp b/src/Yolo3Detection.cpp
index e9b0064..b94eea9 100644
--- a/src/Yolo3Detection.cpp
+++ b/src/Yolo3Detection.cpp
@@ -32,6 +32,9 @@ bool Yolo3Detection::init(const std::string& tensor_path, const int n_classes, c
         memcpy(yolo[i]->bias_h, yRT->bias, sizeof(dnnType)*num*nMasks*2);
         yolo[i]->input_dim = yolo[i]->output_dim = tk::dnn::dataDim_t(1, yRT->c, yRT->h, yRT->w);
         yolo[i]->classesNames = yRT->classesNames;
+        yolo[i]->nms_thresh = yRT->nms_thresh;
+        yolo[i]->nsm_kind = (tk::dnn::Yolo::nmsKind_t) yRT->nms_kind;
+        yolo[i]->new_coords = yRT->new_coords;
     }
 
     dets = tk::dnn::Yolo::allocateDetections(tk::dnn::Yolo::MAX_DETECTIONS, classes);
@@ -102,9 +105,9 @@ void Yolo3Detection::postprocess(const int bi, const bool mAP){
     nDets = 0;
     for(int i=0; i<netRT->pluginFactory->n_yolos; i++) {
         yolo[i]->dstData = rt_out[i];
-        yolo[i]->computeDetections(dets, nDets, netRT->input_dim.w, netRT->input_dim.h, confThreshold);
+        yolo[i]->computeDetections(dets, nDets, netRT->input_dim.w, netRT->input_dim.h, confThreshold, yolo[i]->new_coords);
     }
-    tk::dnn::Yolo::mergeDetections(dets, nDets, classes);
+    tk::dnn::Yolo::mergeDetections(dets, nDets, classes, yolo[0]->nms_thresh, yolo[0]->nsm_kind);
 
     // fill detected
     detected.clear();
diff --git a/tests/darknet/cfg/yolo4x.cfg b/tests/darknet/cfg/yolo4x.cfg
new file mode 100644
index 0000000..89f2564
--- /dev/null
+++ b/tests/darknet/cfg/yolo4x.cfg
@@ -0,0 +1,1427 @@
+[net]
+# Testing
+#batch=1
+#subdivisions=1
+# Training
+batch=64
+subdivisions=8
+width=672
+height=672
+channels=3
+momentum=0.949
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+learning_rate=0.00261
+burn_in=1000
+max_batches = 500500
+policy=steps
+steps=400000,450000
+scales=.1,.1
+
+mosaic=1
+
+letter_box=1
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=40
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=80
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-13
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-34
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-34
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=1280
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-19
+
+[convolutional]
+batch_normalize=1
+filters=1280
+size=1
+stride=1
+pad=1
+activation=mish
+
+########################## 6 0 6 6 3
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=640
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+### SPP ###
+[maxpool]
+stride=1
+size=5
+
+[route]
+layers=-2
+
+[maxpool]
+stride=1
+size=9
+
+[route]
+layers=-4
+
+[maxpool]
+stride=1
+size=13
+
+[route]
+layers=-1,-3,-5,-6
+### End SPP ###
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=640
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=640
+activation=mish
+
+[route]
+layers = -1, -15
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[upsample]
+stride=2
+
+[route]
+layers = 94
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1, -3
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=320
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=320
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=320
+activation=mish
+
+[route]
+layers = -1, -8
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[upsample]
+stride=2
+
+[route]
+layers = 57
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1, -3
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=160
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=160
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=160
+activation=mish
+
+[route]
+layers = -1, -8
+
+[convolutional]
+batch_normalize=1
+filters=160
+size=1
+stride=1
+pad=1
+activation=mish
+
+##########################
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=320
+activation=mish
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=linear
+
+
+[yolo]
+mask = 0,1,2
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=80
+num=9
+jitter=.1
+objectness_smooth=0
+ignore_thresh = .7
+truth_thresh = 1
+#random=1
+resize=1.5
+iou_thresh=0.2
+iou_normalizer=0.05
+cls_normalizer=0.5
+obj_normalizer=4.0
+iou_loss=ciou
+nms_kind=diounms
+beta_nms=0.6
+new_coords=1
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=2
+pad=1
+filters=320
+activation=mish
+
+[route]
+layers = -1, -22
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=320
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=320
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=320
+activation=mish
+
+[route]
+layers = -1,-8
+
+[convolutional]
+batch_normalize=1
+filters=320
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=640
+activation=mish
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=linear
+
+
+[yolo]
+mask = 3,4,5
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=80
+num=9
+jitter=.1
+objectness_smooth=1
+ignore_thresh = .7
+truth_thresh = 1
+#random=1
+resize=1.5
+iou_thresh=0.2
+iou_normalizer=0.05
+cls_normalizer=0.5
+obj_normalizer=1.0
+iou_loss=ciou
+nms_kind=diounms
+beta_nms=0.6
+new_coords=1
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=2
+pad=1
+filters=640
+activation=mish
+
+[route]
+layers = -1, -55
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=640
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=640
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=640
+activation=mish
+
+[route]
+layers = -1,-8
+
+[convolutional]
+batch_normalize=1
+filters=640
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1280
+activation=mish
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=linear
+
+
+[yolo]
+mask = 6,7,8
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=80
+num=9
+jitter=.1
+objectness_smooth=1
+ignore_thresh = .7
+truth_thresh = 1
+#random=1
+resize=1.5
+iou_thresh=0.2
+iou_normalizer=0.05
+cls_normalizer=0.5
+obj_normalizer=0.4
+iou_loss=ciou
+nms_kind=diounms
+beta_nms=0.6
+new_coords=1
diff --git a/tests/darknet/yolo4x.cpp b/tests/darknet/yolo4x.cpp
new file mode 100644
index 0000000..b9ad003
--- /dev/null
+++ b/tests/darknet/yolo4x.cpp
@@ -0,0 +1,36 @@
+#include<iostream>
+#include<vector>
+#include "tkdnn.h"
+#include "test.h"
+#include "DarknetParser.h"
+
+int main() {
+    std::string bin_path  = "yolo4x";
+    std::vector<std::string> input_bins = { 
+        bin_path + "/layers/input.bin"
+    };
+    std::vector<std::string> output_bins = {
+        bin_path + "/debug/layer168_out.bin",
+        bin_path + "/debug/layer185_out.bin",
+        bin_path + "/debug/layer202_out.bin"
+    };
+    std::string wgs_path  = bin_path + "/layers";
+    std::string cfg_path  = std::string(TKDNN_PATH) + "/tests/darknet/cfg/yolo4x.cfg";
+    std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/coco.names";
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/BLPpiAigZJLorQD/download");
+
+
+
+    // parse darknet network
+    tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
+    net->print();
+
+    //convert network to tensorRT
+    tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
+    
+    int ret = testInference(input_bins, output_bins, net, netRT);
+    net->releaseLayers();
+    delete net;
+    delete netRT;
+    return ret;
+}

From b8855b9599e52a51b371e99255063cd6f00fecd7 Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Mon, 23 Nov 2020 11:34:06 +0100
Subject: [PATCH 22/53] Update README

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 README.md | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/README.md b/README.md
index cdfa25b..84e0037 100644
--- a/README.md
+++ b/README.md
@@ -352,7 +352,8 @@ This demo also creates a json file named ```net_name_COCO_res.json``` containing
 | csresnext50-panet-spp    | Cross Stage Partial Network <sup>7</sup>     | [COCO 2014](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/Kcs4xBozwY4wFx8/download)     |
 | yolo4             | Yolov4 <sup>8</sup>                           | [COCO 2017](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/d97CFzYqCPCp5Hg/download)     |
 | yolo4_berkeley             | Yolov4 <sup>8</sup>                           | [BDD100K  ](https://bair.berkeley.edu/blog/2018/05/30/bdd/)                          | 10        | 540x320       | [weights](https://cloud.hipert.unimore.it/s/nkWFa5fgb4NTdnB/download)     |
-| yolo4tiny             | Yolov4 tiny                           | [COCO 2017](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/iRnc4pSqmx78gJs/download)     |
+| yolo4tiny             | Yolov4 tiny <sup>9</sup>                           | [COCO 2017](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/iRnc4pSqmx78gJs/download)     |
+| yolo4x             | Yolov4x-mish  <sup>9</sup>                          | [COCO 2017](http://cocodataset.org/)                          | 80        | 672x672       | [weights](https://cloud.hipert.unimore.it/s/BLPpiAigZJLorQD/download)     |
 
 
 ## References
@@ -365,3 +366,4 @@ This demo also creates a json file named ```net_name_COCO_res.json``` containing
 6. He, Kaiming, et al. "Deep residual learning for image recognition." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
 7. Wang, Chien-Yao, et al. "CSPNet: A New Backbone that can Enhance Learning Capability of CNN." arXiv preprint arXiv:1911.11929 (2019).
 8. Bochkovskiy, Alexey, Chien-Yao Wang, and Hong-Yuan Mark Liao. "YOLOv4: Optimal Speed and Accuracy of Object Detection." arXiv preprint arXiv:2004.10934 (2020).
+9. Bochkovskiy, Alexey, "Yolo v4, v3 and v2 for Windows and Linux" (https://github.com/AlexeyAB/darknet)

From 56feb54377c0678e42077fabbf84ce2fc138f4c5 Mon Sep 17 00:00:00 2001
From: perseusdg <f20180523@goa.bits-pilani.ac.in>
Date: Thu, 21 Jan 2021 00:22:15 +0400
Subject: [PATCH 23/53] able to build kernels as shared object file(dll),and
 minor changes to lstm.cpp and utils.cpp to overcome minor msvc build errors

---
 CMakeLists.txt                  | 11 +++++++++--
 include/tkDNN/DetectionNN.h     |  5 +++++
 include/tkDNN/ImuOdom.h         |  6 ++++++
 include/tkDNN/Int8BatchStream.h |  7 ++++++-
 include/tkDNN/utils.h           |  8 ++++++++
 src/LSTM.cpp                    | 13 +++++++++----
 src/utils.cpp                   |  7 ++++++-
 7 files changed, 49 insertions(+), 8 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 8c8619d..03e26c5 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,7 +2,13 @@ cmake_minimum_required(VERSION 3.5)
 
 project (tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
+if(LINUX)
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
+endif()
+if(WIN32)
+set(CMAKE_CXX_STANDARD 14)
+set(CMAKE_CXX_FLAGS "/O2 /FS ")
+endif()
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include/tkDNN)
 
 # project specific flags
@@ -18,7 +24,7 @@ add_definitions(-DTKDNN_PATH="${CMAKE_CURRENT_SOURCE_DIR}")
 find_package(CUDA 9.0 REQUIRED)
 SET(CUDA_SEPARABLE_COMPILATION ON)
 #set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -arch=sm_30 --compiler-options '-fPIC'")
-set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32)
+set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -arch=sm_61 )
 
 find_package(CUDNN REQUIRED)
 include_directories(${CUDNN_INCLUDE_DIR})
@@ -28,6 +34,7 @@ include_directories(${CUDNN_INCLUDE_DIR})
 file(GLOB tkdnn_CUSRC "src/kernels/*.cu" "src/sorting.cu")
 cuda_include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include ${CUDA_INCLUDE_DIRS} ${CUDNN_INCLUDE_DIRS})
 cuda_add_library(kernels SHARED ${tkdnn_CUSRC})
+target_link_libraries(kernels ${CUDA_CUBLAS_LIBRARIES})
 
 
 #-------------------------------------------------------------------------------
@@ -48,7 +55,7 @@ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DOPENCV")
 file(GLOB tkdnn_SRC "src/*.cpp")
 set(tkdnn_LIBS kernels ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} ${CUDNN_LIBRARIES} ${OpenCV_LIBS} yaml-cpp)
 
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include ${CUDA_INCLUDE_DIRS} ${OPENCV_INCLUDE_DIRS} ${NVINFER_INCLUDES})
 add_library(tkDNN SHARED ${tkdnn_SRC})
 target_link_libraries(tkDNN ${tkdnn_LIBS})
diff --git a/include/tkDNN/DetectionNN.h b/include/tkDNN/DetectionNN.h
index 0498d41..9ce33c1 100644
--- a/include/tkDNN/DetectionNN.h
+++ b/include/tkDNN/DetectionNN.h
@@ -4,7 +4,12 @@
 #include <iostream>
 #include <signal.h>
 #include <stdlib.h>    
+#ifdef __linux__
 #include <unistd.h>
+#elif _WIN32
+#include <Windows.h>
+#endif 
+
 #include <mutex>
 #include "utils.h"
 
diff --git a/include/tkDNN/ImuOdom.h b/include/tkDNN/ImuOdom.h
index 6d8d4cb..aace012 100644
--- a/include/tkDNN/ImuOdom.h
+++ b/include/tkDNN/ImuOdom.h
@@ -1,7 +1,13 @@
 #include <iostream>
 #include <signal.h>
 #include <stdlib.h>     /* srand, rand */
+
+#ifdef __linux__
 #include <unistd.h>
+#elif _WIN32
+#include <Windows.h>
+#endif 
+
 #include <mutex>
 #include <Eigen/Dense>
 #include "utils.h"
diff --git a/include/tkDNN/Int8BatchStream.h b/include/tkDNN/Int8BatchStream.h
index 4349c1f..7d2cef5 100644
--- a/include/tkDNN/Int8BatchStream.h
+++ b/include/tkDNN/Int8BatchStream.h
@@ -11,8 +11,13 @@
 #include <fstream>
 #include <iomanip>
 #include <signal.h>
-#include <stdlib.h>    
+#include <stdlib.h>
+#ifdef __linux__    
 #include <unistd.h>
+#elif _WIN32
+#include <Windows.h>
+#endif
+
 #include <mutex>
 
 #include "NvInfer.h"
diff --git a/include/tkDNN/utils.h b/include/tkDNN/utils.h
index 538a3f3..cb3c18f 100644
--- a/include/tkDNN/utils.h
+++ b/include/tkDNN/utils.h
@@ -12,7 +12,12 @@
 #include <cublas_v2.h>
 #include <cudnn.h>
 
+#ifdef __linux__
 #include <unistd.h>
+#elif _WIN32
+#include <Windows.h>
+#endif 
+
 #include <ios>
 
 
@@ -39,6 +44,7 @@
 #define TKDNN_VERBOSE 0
 
 // Simple Timer 
+#ifdef __linux__
 #define TKDNN_TSTART timespec start, end;                               \
                     clock_gettime(CLOCK_MONOTONIC, &start);            
 
@@ -48,6 +54,8 @@
     if(show) std::cout<<col<<"Time:"<<std::setw(16)<<t_ns<<" ms\n"<<COL_END; 
 
 #define TKDNN_TSTOP TKDNN_TSTOP_C(COL_CYANB, TKDNN_VERBOSE)
+#elif _WIN32
+#endif 
 
 /********************************************************
  * Prints the error message, and exits
diff --git a/src/LSTM.cpp b/src/LSTM.cpp
index 7b87711..d0429e0 100644
--- a/src/LSTM.cpp
+++ b/src/LSTM.cpp
@@ -87,17 +87,22 @@ LSTM::LSTM( Network *net, int hiddensize, bool returnSeq, std::string fname_weig
     checkCUDNN(cudnnCreateRNNDescriptor(&rnnDesc));
 
 #if CUDNN_MAJOR > 7
-    checkCUDNN(cudnnSetRNNDescriptor_v6(net->cudnnHandle,
+    checkCUDNN(cudnnSetRNNDescriptor_v6(net->cudnnHandle,rnnDesc, stateSize, numLayers, dropoutDesc,
+                                        cudnnRNNInputMode_t::CUDNN_LINEAR_INPUT,
+            //(bidirectional ? cudnnDirectionMode_t::CUDNN_BIDIRECTIONAL : cudnnDirectionMode_t::CUDNN_UNIDIRECTIONAL),
+                                        cudnnDirectionMode_t::CUDNN_UNIDIRECTIONAL,
+                                        cudnnRNNMode_t::CUDNN_LSTM,
+                                        cudnnRNNAlgo_t::CUDNN_RNN_ALGO_STANDARD,
+                                        net->dataType));
 #else
-    checkCUDNN(cudnnSetRNNDescriptor(net->cudnnHandle,
-#endif
-        rnnDesc, stateSize, numLayers, dropoutDesc,
+    checkCUDNN(cudnnSetRNNDescriptor(net->cudnnHandle,rnnDesc, stateSize, numLayers, dropoutDesc,
         cudnnRNNInputMode_t::CUDNN_LINEAR_INPUT,
         //(bidirectional ? cudnnDirectionMode_t::CUDNN_BIDIRECTIONAL : cudnnDirectionMode_t::CUDNN_UNIDIRECTIONAL),
         cudnnDirectionMode_t::CUDNN_UNIDIRECTIONAL,
         cudnnRNNMode_t::CUDNN_LSTM,
         cudnnRNNAlgo_t::CUDNN_RNN_ALGO_STANDARD,
         net->dataType));
+#endif
 
 
     // Get temp space sizes
diff --git a/src/utils.cpp b/src/utils.cpp
index 65030f0..e143cce 100644
--- a/src/utils.cpp
+++ b/src/utils.cpp
@@ -170,6 +170,7 @@ void getMemUsage(double& vm_usage_kb, double& resident_set_kb){
    using std::ios_base;
    using std::ifstream;
    using std::string;
+   SYSTEM_INFO sysInfo;
 
    vm_usage_kb     = 0.0;
    resident_set_kb = 0.0;
@@ -191,8 +192,12 @@ void getMemUsage(double& vm_usage_kb, double& resident_set_kb){
                >> O >> itrealvalue >> starttime >> vsize >> rss;
 
    stat_stream.close();
-
+#ifdef __linux__
    long page_size_kb = sysconf(_SC_PAGE_SIZE) / 1024; // in case x86-64 is configured to use 2MB pages
+#elif _WIN32
+ long page_size_kb = sysInfo.dwPageSize/1024;
+#endif
+
    vm_usage_kb     = vsize / 1024.0;  
    resident_set_kb = rss * page_size_kb;
 }

From 512acd8cba99c7ec21d542b32a2881671da39cba Mon Sep 17 00:00:00 2001
From: perseusdg <f20180523@goa.bits-pilani.ac.in>
Date: Thu, 21 Jan 2021 09:29:27 +0400
Subject: [PATCH 24/53] minor fixes

---
 Issues.md             |  1 +
 demo/demo/map.cpp     |  5 +++++
 include/tkDNN/utils.h | 14 +++++++++++---
 src/Yolo.cpp          |  9 +++++----
 4 files changed, 22 insertions(+), 7 deletions(-)
 create mode 100644 Issues.md

diff --git a/Issues.md b/Issues.md
new file mode 100644
index 0000000..4875b13
--- /dev/null
+++ b/Issues.md
@@ -0,0 +1 @@
+1)error C2131 @ Yolo3Detection.cpp(97) -> expression doesnt evaluate to a constant caused to read of variable outside its lifetime
\ No newline at end of file
diff --git a/demo/demo/map.cpp b/demo/demo/map.cpp
index 356e35a..8e363ee 100644
--- a/demo/demo/map.cpp
+++ b/demo/demo/map.cpp
@@ -2,7 +2,12 @@
 #include <iostream>
 #include <signal.h>
 #include <stdlib.h>     /* srand, rand */
+#ifdef __linux__
 #include <unistd.h>
+#elif _WIN32
+#include <Windows.h>
+#endif
+
 #include <mutex>
 #include "utils.h"
 
diff --git a/include/tkDNN/utils.h b/include/tkDNN/utils.h
index cb3c18f..1404509 100644
--- a/include/tkDNN/utils.h
+++ b/include/tkDNN/utils.h
@@ -15,10 +15,12 @@
 #ifdef __linux__
 #include <unistd.h>
 #elif _WIN32
-#include <Windows.h>
-#endif 
+#define NOMINMAX
+#include <windows.h>
+#endif
 
 #include <ios>
+#include <chrono>
 
 
 #define dnnType float
@@ -55,7 +57,13 @@
 
 #define TKDNN_TSTOP TKDNN_TSTOP_C(COL_CYANB, TKDNN_VERBOSE)
 #elif _WIN32
-#endif 
+#define TKDNN_TSTART auto start = std::chrono::high_resolution_clock::now();
+#define TKDNN_TSTOP auto stop = std::chrono::high_resolution_clock::now(); \
+std::chrono::duration<double> duration = stop -start;                      \
+auto time_ms = std::chrono::duration_cast<std::chrono::microseconds>(duration);\
+double t_ns = time_ms.count();
+#endif
+
 
 /********************************************************
  * Prints the error message, and exits
diff --git a/src/Yolo.cpp b/src/Yolo.cpp
index 9737e74..6d0b546 100644
--- a/src/Yolo.cpp
+++ b/src/Yolo.cpp
@@ -9,6 +9,7 @@
 #include "Layer.h"
 #include "kernels.h"
 
+
 namespace tk { namespace dnn {
 
 Yolo::Yolo(Network *net, int classes, int num, std::string fname_weights, int n_masks, float scale_xy, double nms_thresh, nmsKind_t nsm_kind, int new_coords) : 
@@ -209,10 +210,10 @@ float yolo_box_iou(Yolo::box a, Yolo::box b)
 }
 
 void box_c(const Yolo::box a, const Yolo::box b, float& top, float& bot, float& left, float& right) {
-    top = std::min(a.y - a.h / 2, b.y - b.h / 2);
-    bot = std::max(a.y + a.h / 2, b.y + b.h / 2);
-    left = std::min(a.x - a.w / 2, b.x - b.w / 2);
-    right = std::max(a.x + a.w / 2, b.x + b.w / 2);
+    top = (std::min)(a.y - a.h / 2, b.y - b.h / 2);
+    bot = (std::max)(a.y + a.h / 2, b.y + b.h / 2);
+    left = (std::min)(a.x - a.w / 2, b.x - b.w / 2);
+    right = (std::max)(a.x + a.w / 2, b.x + b.w / 2);
 }
 
 // https://github.com/Zzh-tju/DIoU-darknet

From adac8576b0faf515ad3f459b1f50fd16cef6d64d Mon Sep 17 00:00:00 2001
From: Micaela Verucchi <micaelaverucchi@gmail.com>
Date: Fri, 22 Jan 2021 17:57:42 +0100
Subject: [PATCH 25/53] Add support to Scaled-YOLO4, update Yolov4x-mish
 (tested)

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
---
 README.md                                     |    4 +-
 include/tkDNN/Layer.h                         |    7 +-
 include/tkDNN/NetworkRT.h                     |    1 +
 .../tkDNN/pluginsRT/ActivationLogisticRT.h    |   60 +
 include/tkDNN/pluginsRT/YoloRT.h              |   14 +-
 scripts/test_all_tests.sh                     |    5 +-
 src/Activation.cpp                            |    4 +
 src/DarknetParser.cpp                         |    1 +
 src/NetworkRT.cpp                             |   13 +-
 src/Yolo.cpp                                  |   21 +-
 tests/darknet/cfg/yolo4-csp.cfg               | 1279 +++++++++++++++++
 tests/darknet/cfg/yolo4x.cfg                  |   21 +-
 tests/darknet/yolo4-csp.cpp                   |   36 +
 tests/darknet/yolo4x.cpp                      |    2 +-
 14 files changed, 1441 insertions(+), 27 deletions(-)
 create mode 100644 include/tkDNN/pluginsRT/ActivationLogisticRT.h
 create mode 100644 tests/darknet/cfg/yolo4-csp.cfg
 create mode 100644 tests/darknet/yolo4-csp.cpp

diff --git a/README.md b/README.md
index 84e0037..d9927c0 100644
--- a/README.md
+++ b/README.md
@@ -353,7 +353,8 @@ This demo also creates a json file named ```net_name_COCO_res.json``` containing
 | yolo4             | Yolov4 <sup>8</sup>                           | [COCO 2017](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/d97CFzYqCPCp5Hg/download)     |
 | yolo4_berkeley             | Yolov4 <sup>8</sup>                           | [BDD100K  ](https://bair.berkeley.edu/blog/2018/05/30/bdd/)                          | 10        | 540x320       | [weights](https://cloud.hipert.unimore.it/s/nkWFa5fgb4NTdnB/download)     |
 | yolo4tiny             | Yolov4 tiny <sup>9</sup>                           | [COCO 2017](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/iRnc4pSqmx78gJs/download)     |
-| yolo4x             | Yolov4x-mish  <sup>9</sup>                          | [COCO 2017](http://cocodataset.org/)                          | 80        | 672x672       | [weights](https://cloud.hipert.unimore.it/s/BLPpiAigZJLorQD/download)     |
+| yolo4x             | Yolov4x-mish  <sup>9</sup>                          | [COCO 2017](http://cocodataset.org/)                          | 80        | 640x640       | [weights](https://cloud.hipert.unimore.it/s/5MFjtNtgbDGdJEo/download)     |
+| yolo4x-cps            | Scaled Yolov4 <sup>10</sup>                          | [COCO 2017](http://cocodataset.org/)                          | 80        | 512x512       | [weights](https://cloud.hipert.unimore.it/s/AfzHE4BfTeEm2gH/download)     |
 
 
 ## References
@@ -367,3 +368,4 @@ This demo also creates a json file named ```net_name_COCO_res.json``` containing
 7. Wang, Chien-Yao, et al. "CSPNet: A New Backbone that can Enhance Learning Capability of CNN." arXiv preprint arXiv:1911.11929 (2019).
 8. Bochkovskiy, Alexey, Chien-Yao Wang, and Hong-Yuan Mark Liao. "YOLOv4: Optimal Speed and Accuracy of Object Detection." arXiv preprint arXiv:2004.10934 (2020).
 9. Bochkovskiy, Alexey, "Yolo v4, v3 and v2 for Windows and Linux" (https://github.com/AlexeyAB/darknet)
+10. Wang, Chien-Yao, Alexey Bochkovskiy, and Hong-Yuan Mark Liao. "Scaled-YOLOv4: Scaling Cross Stage Partial Network." arXiv preprint arXiv:2011.08036 (2020).
diff --git a/include/tkDNN/Layer.h b/include/tkDNN/Layer.h
index 25c4565..e097372 100644
--- a/include/tkDNN/Layer.h
+++ b/include/tkDNN/Layer.h
@@ -19,6 +19,7 @@ enum layerType_t {
     LAYER_ACTIVATION_CRELU,
     LAYER_ACTIVATION_LEAKY,
     LAYER_ACTIVATION_MISH,
+    LAYER_ACTIVATION_LOGISTIC,
     LAYER_FLATTEN,
     LAYER_RESHAPE,
     LAYER_MULADD,
@@ -68,6 +69,7 @@ public:
             case LAYER_ACTIVATION_CRELU:    return "ActivationCReLU";
             case LAYER_ACTIVATION_LEAKY:    return "ActivationLeaky";
             case LAYER_ACTIVATION_MISH:     return "ActivationMish";
+            case LAYER_ACTIVATION_LOGISTIC: return "ActivationLogistic";
             case LAYER_FLATTEN:             return "Flatten";
             case LAYER_RESHAPE:             return "Reshape";
             case LAYER_MULADD:              return "MulAdd";
@@ -212,7 +214,8 @@ public:
 typedef enum {
     ACTIVATION_ELU     = 100,
     ACTIVATION_LEAKY   = 101,
-    ACTIVATION_MISH   = 102
+    ACTIVATION_MISH   = 102,
+    ACTIVATION_LOGISTIC   = 103
 } tkdnnActivationMode_t;
 
 /**
@@ -233,6 +236,8 @@ public:
             return LAYER_ACTIVATION_LEAKY;
         else if (act_mode == ACTIVATION_MISH)
             return LAYER_ACTIVATION_MISH;
+        else if (act_mode == ACTIVATION_LOGISTIC)
+            return LAYER_ACTIVATION_LOGISTIC;
         else
             return LAYER_ACTIVATION;
          };
diff --git a/include/tkDNN/NetworkRT.h b/include/tkDNN/NetworkRT.h
index 4c6c816..4fe2e0e 100644
--- a/include/tkDNN/NetworkRT.h
+++ b/include/tkDNN/NetworkRT.h
@@ -24,6 +24,7 @@ template<typename T> T readBUF(const char*& buffer)
 
 using namespace nvinfer1;
 #include "pluginsRT/ActivationLeakyRT.h"
+#include "pluginsRT/ActivationLogisticRT.h"
 #include "pluginsRT/ActivationReLUCeilingRT.h"
 #include "pluginsRT/ActivationMishRT.h"
 #include "pluginsRT/ReorgRT.h"
diff --git a/include/tkDNN/pluginsRT/ActivationLogisticRT.h b/include/tkDNN/pluginsRT/ActivationLogisticRT.h
new file mode 100644
index 0000000..a1ceb6b
--- /dev/null
+++ b/include/tkDNN/pluginsRT/ActivationLogisticRT.h
@@ -0,0 +1,60 @@
+#include<cassert>
+#include "../kernels.h"
+
+class ActivationLogisticRT : public IPlugin {
+
+public:
+	ActivationLogisticRT() {
+
+
+	}
+
+	~ActivationLogisticRT(){
+
+	}
+
+	int getNbOutputs() const override {
+		return 1;
+	}
+
+	Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) override {
+		return inputs[0];
+	}
+
+	void configure(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, int maxBatchSize) override {
+		size = 1;
+		for(int i=0; i<outputDims[0].nbDims; i++)
+			size *= outputDims[0].d[i];
+	}
+
+	int initialize() override {
+
+		return 0;
+	}
+
+	virtual void terminate() override {
+	}
+
+	virtual size_t getWorkspaceSize(int maxBatchSize) const override {
+		return 0;
+	}
+
+	virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override {
+
+		activationLOGISTICForward((dnnType*)reinterpret_cast<const dnnType*>(inputs[0]), 
+											reinterpret_cast<dnnType*>(outputs[0]), batchSize*size, stream);
+		return 0;
+	}
+
+
+	virtual size_t getSerializationSize() override {
+		return 1*sizeof(int);
+	}
+
+	virtual void serialize(void* buffer) override {
+		char *buf = reinterpret_cast<char*>(buffer);
+		tk::dnn::writeBUF(buf, size);
+	}
+
+	int size;
+};
diff --git a/include/tkDNN/pluginsRT/YoloRT.h b/include/tkDNN/pluginsRT/YoloRT.h
index 9af8587..f2bdaa1 100644
--- a/include/tkDNN/pluginsRT/YoloRT.h
+++ b/include/tkDNN/pluginsRT/YoloRT.h
@@ -67,15 +67,17 @@ public:
 		for (int b = 0; b < batchSize; ++b){
 			for(int n = 0; n < n_masks; ++n){
 				int index = entry_index(b, n*w*h, 0);
-				if (new_coords == 1)
-					activationLOGISTICForward(srcData + index, dstData + index, 4*w*h, stream); //x,y,w,h
-				else
+				if (new_coords == 1){
+					if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
+				}
+				else{
 					activationLOGISTICForward(srcData + index, dstData + index, 2*w*h, stream); //x,y
 
-				if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
+					if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
 				
-				index = entry_index(b, n*w*h, 4);
-				activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*w*h, stream);
+					index = entry_index(b, n*w*h, 4);
+					activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*w*h, stream);
+				}
 			}
 		}
 
diff --git a/scripts/test_all_tests.sh b/scripts/test_all_tests.sh
index af04aff..6aab775 100644
--- a/scripts/test_all_tests.sh
+++ b/scripts/test_all_tests.sh
@@ -69,10 +69,11 @@ do
     echo -e "${ORANGE}Batch $TKDNN_BATCHSIZE ${NC}"
     
     test_net mnist    
-    ./test_imuodom &>> $out_file
-    print_output $? imuodom
+    # ./test_imuodom &>> $out_file
+    # print_output $? imuodom
 
     test_net yolo4
+    test_net yolo4-csp
     test_net yolo4x
     test_net yolo4_berkeley
     test_net yolo4tiny
diff --git a/src/Activation.cpp b/src/Activation.cpp
index 28c7624..a8642e7 100644
--- a/src/Activation.cpp
+++ b/src/Activation.cpp
@@ -52,6 +52,10 @@ dnnType* Activation::infer(dataDim_t &dim, dnnType* srcData) {
     else if(act_mode == ACTIVATION_MISH) {
         activationMishForward(srcData, dstData, dim.tot());
 
+    }
+    else if(act_mode == ACTIVATION_LOGISTIC) {
+        activationLOGISTICForward(srcData, dstData, dim.tot());
+
     } else {
         dnnType alpha = dnnType(1);
         dnnType beta  = dnnType(0);
diff --git a/src/DarknetParser.cpp b/src/DarknetParser.cpp
index 7b5410c..69b6b29 100644
--- a/src/DarknetParser.cpp
+++ b/src/DarknetParser.cpp
@@ -187,6 +187,7 @@ namespace tk { namespace dnn {
             if(f.activation == "relu") act = tkdnnActivationMode_t(CUDNN_ACTIVATION_RELU);
             else if(f.activation == "leaky") act = tk::dnn::ACTIVATION_LEAKY;
             else if(f.activation == "mish") act = tk::dnn::ACTIVATION_MISH;
+            else if(f.activation == "logistic") act = tk::dnn::ACTIVATION_LOGISTIC;
             else { FatalError("activation not supported: " + f.activation); }
             netLayers[netLayers.size()-1] = new tk::dnn::Activation(net, act);
         };
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index 501ade4..c915ba5 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -226,7 +226,7 @@ ILayer* NetworkRT::convert_layer(ITensor *input, Layer *l) {
         return convert_layer(input, (Conv2d*) l);
     if(type == LAYER_POOLING)
         return convert_layer(input, (Pooling*) l);
-    if(type == LAYER_ACTIVATION || type == LAYER_ACTIVATION_CRELU || type == LAYER_ACTIVATION_LEAKY || type == LAYER_ACTIVATION_MISH)
+    if(type == LAYER_ACTIVATION || type == LAYER_ACTIVATION_CRELU || type == LAYER_ACTIVATION_LEAKY || type == LAYER_ACTIVATION_MISH || type == LAYER_ACTIVATION_LOGISTIC)
         return convert_layer(input, (Activation*) l);
     if(type == LAYER_SOFTMAX)
         return convert_layer(input, (Softmax*) l);
@@ -421,6 +421,12 @@ ILayer* NetworkRT::convert_layer(ITensor *input, Activation *l) {
         checkNULL(lRT);
         return lRT;
     }
+    else if(l->act_mode == ACTIVATION_LOGISTIC) {
+        IPlugin *plugin = new ActivationLogisticRT();
+        IPluginLayer *lRT = networkRT->addPlugin(&input, 1, *plugin);
+        checkNULL(lRT);
+        return lRT;
+    }
     else {
         FatalError("this Activation mode is not yet implemented");
         return NULL;
@@ -653,6 +659,11 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         a->size = readBUF<int>(buf);
         return a;
     }
+    if(name.find("ActivationLogistic") == 0) {
+        ActivationLogisticRT *a = new ActivationLogisticRT();
+        a->size = readBUF<int>(buf);
+        return a;
+    }
     if(name.find("ActivationCReLU") == 0) {
         ActivationReLUCeiling *a = new ActivationReLUCeiling(readBUF<float>(buf));
         a->size = readBUF<int>(buf);
diff --git a/src/Yolo.cpp b/src/Yolo.cpp
index 9737e74..61ed3cf 100644
--- a/src/Yolo.cpp
+++ b/src/Yolo.cpp
@@ -72,8 +72,8 @@ Yolo::box get_yolo_box(float *x, float *biases, int n, int index, int i, int j,
         b.h = exp(x[index + 3*stride]) * biases[2*n+1] / h;
     }
     else{
-        b.x = (i + x[index + 0 * stride] * 2 - 0.5) / lw;
-        b.y = (j + x[index + 1 * stride] * 2 - 0.5) / lh;
+        b.x = (i + x[index + 0 * stride] ) / lw;
+        b.y = (j + x[index + 1 * stride] ) / lh;
         b.w = x[index + 2 * stride] * x[index + 2 * stride] * 4 * biases[2 * n] / w;
         b.h = x[index + 3 * stride] * x[index + 3 * stride] * 4 * biases[2 * n + 1] / h;
     }
@@ -87,15 +87,18 @@ dnnType* Yolo::infer(dataDim_t &dim, dnnType* srcData) {
     for (int b = 0; b < dim.n; ++b){
         for(int n = 0; n < n_masks; ++n){
             int index = entry_index(b, n*dim.w*dim.h, 0, classes, input_dim, output_dim);
-            if (new_coords == 1)
-                activationLOGISTICForward(srcData + index, dstData + index, 4*dim.w*dim.h);
-            else
+            std::cout<<"new_coords"<<new_coords<<std::endl;
+            if (new_coords == 1){
+                if (this->scaleXY != 1) scalAdd(dstData + index, 2 * dim.w*dim.h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
+            }
+            else{
                 activationLOGISTICForward(srcData + index, dstData + index, 2*dim.w*dim.h);
 
-            if (this->scaleXY != 1) scalAdd(dstData + index, 2 * dim.w*dim.h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
-            
-            index = entry_index(b, n*dim.w*dim.h, 4, classes, input_dim, output_dim);
-            activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*dim.w*dim.h);
+                if (this->scaleXY != 1) scalAdd(dstData + index, 2 * dim.w*dim.h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
+                
+                index = entry_index(b, n*dim.w*dim.h, 4, classes, input_dim, output_dim);
+                activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*dim.w*dim.h);
+            }
         }
     }
 
diff --git a/tests/darknet/cfg/yolo4-csp.cfg b/tests/darknet/cfg/yolo4-csp.cfg
new file mode 100644
index 0000000..691ec03
--- /dev/null
+++ b/tests/darknet/cfg/yolo4-csp.cfg
@@ -0,0 +1,1279 @@
+[net]
+# Testing
+#batch=1
+#subdivisions=1
+# Training
+batch=64
+subdivisions=8
+width=512
+height=512
+channels=3
+momentum=0.949
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+learning_rate=0.001
+burn_in=1000
+max_batches = 500500
+policy=steps
+steps=400000,450000
+scales=.1,.1
+
+mosaic=1
+
+letter_box=1
+
+ema_alpha=0.9998
+
+#optimized_memory=1
+
+#23:104x104 54:52x52 85:26x26 104:13x13 for 416
+
+
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=2
+pad=1
+activation=mish
+
+#[convolutional]
+#batch_normalize=1
+#filters=64
+#size=1
+#stride=1
+#pad=1
+#activation=mish
+
+#[route]
+#layers = -2
+
+#[convolutional]
+#batch_normalize=1
+#filters=64
+#size=1
+#stride=1
+#pad=1
+#activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+#[convolutional]
+#batch_normalize=1
+#filters=64
+#size=1
+#stride=1
+#pad=1
+#activation=mish
+
+#[route]
+#layers = -1,-7
+
+#[convolutional]
+#batch_normalize=1
+#filters=64
+#size=1
+#stride=1
+#pad=1
+#activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-10
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-28
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-28
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-16
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=mish
+
+##########################
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+### SPP ###
+[maxpool]
+stride=1
+size=5
+
+[route]
+layers=-2
+
+[maxpool]
+stride=1
+size=9
+
+[route]
+layers=-4
+
+[maxpool]
+stride=1
+size=13
+
+[route]
+layers=-1,-3,-5,-6
+### End SPP ###
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[route]
+layers = -1, -13
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[upsample]
+stride=2
+
+[route]
+layers = 79
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1, -3
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[route]
+layers = -1, -6
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[upsample]
+stride=2
+
+[route]
+layers = 48
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1, -3
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=128
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=128
+activation=mish
+
+[route]
+layers = -1, -6
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+##########################
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=logistic
+
+
+[yolo]
+mask = 0,1,2
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=80
+num=9
+jitter=.1
+scale_x_y = 2.0
+objectness_smooth=0
+ignore_thresh = .7
+truth_thresh = 1
+#random=1
+resize=1.5
+iou_thresh=0.2
+iou_normalizer=0.05
+cls_normalizer=0.5
+obj_normalizer=4.0
+iou_loss=ciou
+nms_kind=diounms
+beta_nms=0.6
+new_coords=1
+max_delta=5
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=2
+pad=1
+filters=256
+activation=mish
+
+[route]
+layers = -1, -20
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[route]
+layers = -1,-6
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=logistic
+
+
+[yolo]
+mask = 3,4,5
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=80
+num=9
+jitter=.1
+scale_x_y = 2.0
+objectness_smooth=1
+ignore_thresh = .7
+truth_thresh = 1
+#random=1
+resize=1.5
+iou_thresh=0.2
+iou_normalizer=0.05
+cls_normalizer=0.5
+obj_normalizer=1.0
+iou_loss=ciou
+nms_kind=diounms
+beta_nms=0.6
+new_coords=1
+max_delta=5
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=2
+pad=1
+filters=512
+activation=mish
+
+[route]
+layers = -1, -49
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[route]
+layers = -1,-6
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=mish
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=logistic
+
+
+[yolo]
+mask = 6,7,8
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=80
+num=9
+jitter=.1
+scale_x_y = 2.0
+objectness_smooth=1
+ignore_thresh = .7
+truth_thresh = 1
+#random=1
+resize=1.5
+iou_thresh=0.2
+iou_normalizer=0.05
+cls_normalizer=0.5
+obj_normalizer=0.4
+iou_loss=ciou
+nms_kind=diounms
+beta_nms=0.6
+new_coords=1
+max_delta=2
diff --git a/tests/darknet/cfg/yolo4x.cfg b/tests/darknet/cfg/yolo4x.cfg
index 89f2564..2ff854f 100644
--- a/tests/darknet/cfg/yolo4x.cfg
+++ b/tests/darknet/cfg/yolo4x.cfg
@@ -5,8 +5,8 @@
 # Training
 batch=64
 subdivisions=8
-width=672
-height=672
+width=640
+height=640
 channels=3
 momentum=0.949
 decay=0.0005
@@ -15,7 +15,7 @@ saturation = 1.5
 exposure = 1.5
 hue=.1
 
-learning_rate=0.00261
+learning_rate=0.001
 burn_in=1000
 max_batches = 500500
 policy=steps
@@ -26,6 +26,8 @@ mosaic=1
 
 letter_box=1
 
+#optimized_memory=1
+
 [convolutional]
 batch_normalize=1
 filters=32
@@ -1131,6 +1133,7 @@ size=1
 stride=1
 pad=1
 activation=mish
+stopbackward=800
 
 ##########################
 
@@ -1147,7 +1150,7 @@ size=1
 stride=1
 pad=1
 filters=255
-activation=linear
+activation=logistic
 
 
 [yolo]
@@ -1156,6 +1159,7 @@ anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 4
 classes=80
 num=9
 jitter=.1
+scale_x_y = 2.0
 objectness_smooth=0
 ignore_thresh = .7
 truth_thresh = 1
@@ -1169,6 +1173,7 @@ iou_loss=ciou
 nms_kind=diounms
 beta_nms=0.6
 new_coords=1
+max_delta=5
 
 [route]
 layers = -4
@@ -1275,7 +1280,7 @@ size=1
 stride=1
 pad=1
 filters=255
-activation=linear
+activation=logistic
 
 
 [yolo]
@@ -1284,6 +1289,7 @@ anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 4
 classes=80
 num=9
 jitter=.1
+scale_x_y = 2.0
 objectness_smooth=1
 ignore_thresh = .7
 truth_thresh = 1
@@ -1297,6 +1303,7 @@ iou_loss=ciou
 nms_kind=diounms
 beta_nms=0.6
 new_coords=1
+max_delta=5
 
 [route]
 layers = -4
@@ -1403,7 +1410,7 @@ size=1
 stride=1
 pad=1
 filters=255
-activation=linear
+activation=logistic
 
 
 [yolo]
@@ -1412,6 +1419,7 @@ anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 4
 classes=80
 num=9
 jitter=.1
+scale_x_y = 2.0
 objectness_smooth=1
 ignore_thresh = .7
 truth_thresh = 1
@@ -1425,3 +1433,4 @@ iou_loss=ciou
 nms_kind=diounms
 beta_nms=0.6
 new_coords=1
+max_delta=2
diff --git a/tests/darknet/yolo4-csp.cpp b/tests/darknet/yolo4-csp.cpp
new file mode 100644
index 0000000..af8a7fc
--- /dev/null
+++ b/tests/darknet/yolo4-csp.cpp
@@ -0,0 +1,36 @@
+#include<iostream>
+#include<vector>
+#include "tkdnn.h"
+#include "test.h"
+#include "DarknetParser.h"
+
+int main() {
+    std::string bin_path  = "yolo4-csp";
+    std::vector<std::string> input_bins = { 
+        bin_path + "/layers/input.bin"
+    };
+    std::vector<std::string> output_bins = {
+        bin_path + "/debug/layer144_out.bin",
+        bin_path + "/debug/layer159_out.bin",
+        bin_path + "/debug/layer174_out.bin"
+    };
+    std::string wgs_path  = bin_path + "/layers";
+    std::string cfg_path  = std::string(TKDNN_PATH) + "/tests/darknet/cfg/yolo4-csp.cfg";
+    std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/coco.names";
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/AfzHE4BfTeEm2gH/download");
+
+
+
+    // parse darknet network
+    tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
+    net->print();
+
+    //convert network to tensorRT
+    tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
+    
+    int ret = testInference(input_bins, output_bins, net, netRT);
+    net->releaseLayers();
+    delete net;
+    delete netRT;
+    return ret;
+}
diff --git a/tests/darknet/yolo4x.cpp b/tests/darknet/yolo4x.cpp
index b9ad003..8df1aef 100644
--- a/tests/darknet/yolo4x.cpp
+++ b/tests/darknet/yolo4x.cpp
@@ -17,7 +17,7 @@ int main() {
     std::string wgs_path  = bin_path + "/layers";
     std::string cfg_path  = std::string(TKDNN_PATH) + "/tests/darknet/cfg/yolo4x.cfg";
     std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/coco.names";
-    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/BLPpiAigZJLorQD/download");
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/5MFjtNtgbDGdJEo/download");
 
 
 

From fb52444cdc06b6a3868dd13cdd2cc0d0601cc6c0 Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Mon, 25 Jan 2021 04:12:05 +0530
Subject: [PATCH 26/53] Replaced dynamic arrays with std::vector ,works on
 linux ..needs to be tested on windows after clearing up the lnk2019 error

---
 CMakeLists.txt         | 4 ++--
 src/Yolo3Detection.cpp | 7 ++++---
 src/utils.cpp          | 2 +-
 3 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 03e26c5..c11b56f 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,13 +2,13 @@ cmake_minimum_required(VERSION 3.5)
 
 project (tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
-if(LINUX)
+if(UNIX)
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
 endif()
 if(WIN32)
 set(CMAKE_CXX_STANDARD 14)
 set(CMAKE_CXX_FLAGS "/O2 /FS ")
-endif()
+endif(WIN32)
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include/tkDNN)
 
 # project specific flags
diff --git a/src/Yolo3Detection.cpp b/src/Yolo3Detection.cpp
index b94eea9..0c638e6 100644
--- a/src/Yolo3Detection.cpp
+++ b/src/Yolo3Detection.cpp
@@ -94,9 +94,10 @@ void Yolo3Detection::preprocess(cv::Mat &frame, const int bi){
 void Yolo3Detection::postprocess(const int bi, const bool mAP){
 
     //get yolo outputs
-    dnnType *rt_out[netRT->pluginFactory->n_yolos]; 
-    for(int i=0; i<netRT->pluginFactory->n_yolos; i++) 
-        rt_out[i] = (dnnType*)netRT->buffersRT[i+1] + netRT->buffersDIM[i+1].tot()*bi;
+    std::vector<float *> rt_out;
+    //dnnType *rt_out[netRT->pluginFactory->n_yolos];
+    for(int i=0; i<netRT->pluginFactory->n_yolos; i++)
+        rt_out.push_back((dnnType*)netRT->buffersRT[i+1] + netRT->buffersDIM[i+1].tot()*bi);
 
     float x_ratio =  float(originalSize[bi].width) / float(netRT->input_dim.w);
     float y_ratio =  float(originalSize[bi].height) / float(netRT->input_dim.h);
diff --git a/src/utils.cpp b/src/utils.cpp
index e143cce..bbdf516 100644
--- a/src/utils.cpp
+++ b/src/utils.cpp
@@ -170,7 +170,6 @@ void getMemUsage(double& vm_usage_kb, double& resident_set_kb){
    using std::ios_base;
    using std::ifstream;
    using std::string;
-   SYSTEM_INFO sysInfo;
 
    vm_usage_kb     = 0.0;
    resident_set_kb = 0.0;
@@ -195,6 +194,7 @@ void getMemUsage(double& vm_usage_kb, double& resident_set_kb){
 #ifdef __linux__
    long page_size_kb = sysconf(_SC_PAGE_SIZE) / 1024; // in case x86-64 is configured to use 2MB pages
 #elif _WIN32
+SYSTEM_INFO sysInfo;
  long page_size_kb = sysInfo.dwPageSize/1024;
 #endif
 

From 2d4dececb683ffa28f7f8aaf72a2e645e8c45adb Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Tue, 26 Jan 2021 20:17:43 +0400
Subject: [PATCH 27/53] Builds on windows successfully,issues with
 deserialization and downloading weights

---
 .gitignore           | 3 ++-
 CMakeLists.txt       | 9 +++++----
 demo/demo/demo.cpp   | 2 +-
 include/tkDNN/test.h | 3 ++-
 4 files changed, 10 insertions(+), 7 deletions(-)

diff --git a/.gitignore b/.gitignore
index b56526f..5be5a73 100644
--- a/.gitignore
+++ b/.gitignore
@@ -12,5 +12,6 @@ build/
 *.hdf5
 *.pk
 *.table
+cmake-build-release/
 demo/COCO_val2017
-demo/BDD100K_val
\ No newline at end of file
+demo/BDD100K_val
diff --git a/CMakeLists.txt b/CMakeLists.txt
index c11b56f..4904e5d 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,13 +1,14 @@
 cmake_minimum_required(VERSION 3.5)
 
-project (tkDNN)
+project (tkDNN CUDA)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 if(UNIX)
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
 endif()
 if(WIN32)
 set(CMAKE_CXX_STANDARD 14)
-set(CMAKE_CXX_FLAGS "/O2 /FS ")
+set(CMAKE_CXX_FLAGS "/O2 ")
+set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
 endif(WIN32)
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include/tkDNN)
 
@@ -33,7 +34,7 @@ include_directories(${CUDNN_INCLUDE_DIR})
 # compile
 file(GLOB tkdnn_CUSRC "src/kernels/*.cu" "src/sorting.cu")
 cuda_include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include ${CUDA_INCLUDE_DIRS} ${CUDNN_INCLUDE_DIRS})
-cuda_add_library(kernels SHARED ${tkdnn_CUSRC})
+add_library(kernels SHARED ${tkdnn_CUSRC})
 target_link_libraries(kernels ${CUDA_CUBLAS_LIBRARIES})
 
 
@@ -47,7 +48,7 @@ find_package(OpenCV REQUIRED)
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DOPENCV")
 
 # gives problems in cross-compiling, probably malformed cmake config
-#find_package(yaml-cpp REQUIRED)
+find_package(yaml-cpp REQUIRED)
 
 #-------------------------------------------------------------------------------
 # Build Libraries
diff --git a/demo/demo/demo.cpp b/demo/demo/demo.cpp
index 9f50d0b..609affe 100644
--- a/demo/demo/demo.cpp
+++ b/demo/demo/demo.cpp
@@ -1,7 +1,7 @@
 #include <iostream>
 #include <signal.h>
 #include <stdlib.h>     /* srand, rand */
-#include <unistd.h>
+//#include <unistd.h>
 #include <mutex>
 
 #include "CenternetDetection.h"
diff --git a/include/tkDNN/test.h b/include/tkDNN/test.h
index 13c943e..f1d44bd 100644
--- a/include/tkDNN/test.h
+++ b/include/tkDNN/test.h
@@ -29,7 +29,8 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
     readBinaryFile(input_bins[0], net->input_dim.tot(), &input_h, &data);
 
     // outputs
-    dnnType *cudnn_out[outputs.size()], *rt_out[outputs.size()]; 
+    //dnnType *cudnn_out[outputs.size()], *rt_out[outputs.size()];
+    std::vector<dnnType *> cudnn_out,rt_out;
 
     tk::dnn::dataDim_t dim1 =  net->input_dim; //input dim
     printCenteredTitle(" CUDNN inference ", '=', 30); {

From 4a9031433399b6dbf5d08a6c963f8e3f3a829b72 Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Tue, 26 Jan 2021 23:11:58 +0530
Subject: [PATCH 28/53] minor fixes in test.h

---
 CMakeLists.txt       | 4 ++--
 include/tkDNN/test.h | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 4904e5d..c173eca 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.5)
 
-project (tkDNN CUDA)
+project (tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 if(UNIX)
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
@@ -34,7 +34,7 @@ include_directories(${CUDNN_INCLUDE_DIR})
 # compile
 file(GLOB tkdnn_CUSRC "src/kernels/*.cu" "src/sorting.cu")
 cuda_include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include ${CUDA_INCLUDE_DIRS} ${CUDNN_INCLUDE_DIRS})
-add_library(kernels SHARED ${tkdnn_CUSRC})
+cuda_add_library(kernels SHARED ${tkdnn_CUSRC})
 target_link_libraries(kernels ${CUDA_CUBLAS_LIBRARIES})
 
 
diff --git a/include/tkDNN/test.h b/include/tkDNN/test.h
index f1d44bd..e842269 100644
--- a/include/tkDNN/test.h
+++ b/include/tkDNN/test.h
@@ -40,7 +40,7 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
         TKDNN_TSTOP
         dim1.print();   
     }
-    for(int i=0; i<outputs.size(); i++) cudnn_out[i] = outputs[i]->dstData;
+    for(int i=0; i<outputs.size(); i++) cudnn_out.push_back(outputs[i]->dstData);
 
     if(netRT != nullptr) {
         tk::dnn::dataDim_t dim2 = net->input_dim;
@@ -51,7 +51,7 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
             TKDNN_TSTOP
             dim2.print();
         }
-        for(int i=0; i<outputs.size(); i++) rt_out[i] = (dnnType*)netRT->buffersRT[i+1];
+        for(int i=0; i<outputs.size(); i++) rt_out.push_back((dnnType*)netRT->buffersRT[i+1]);
     }
 
     int ret_cudnn = 0, ret_tensorrt = 0, ret_cudnn_tensorrt = 0; 

From f055341af6cf6a3fb914a95a9aa2561fe3d81df7 Mon Sep 17 00:00:00 2001
From: Ricky Medrano <rickymedrano4@gmail.com>
Date: Tue, 9 Feb 2021 07:57:57 -0800
Subject: [PATCH 29/53] Minor Readme Changes

Added Logistic as a viable activation you can use.
Added conf-thresh as the 7th parameter in the ./demo call.
---
 README.md | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/README.md b/README.md
index d9927c0..5055e99 100644
--- a/README.md
+++ b/README.md
@@ -195,6 +195,7 @@ All models from darknet are now parsed directly from cfg, you still need to expo
   relu
   leaky
   mish
+  logistic
 </details>
 
 ## Run the demo 
@@ -217,7 +218,7 @@ Once you have successfully created your rt file, run the demo:
 ```
 In general the demo program takes 7 parameters:
 ```
-./demo <network-rt-file> <path-to-video> <kind-of-network> <number-of-classes> <n-batches> <show-flag>
+./demo <network-rt-file> <path-to-video> <kind-of-network> <number-of-classes> <n-batches> <show-flag> <conf-thresh>
 ```
 where
 *  ```<network-rt-file>``` is the rt file generated by a test

From 6aa8666be54ce7726654afd3062704555a836161 Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Wed, 10 Mar 2021 12:59:04 +0530
Subject: [PATCH 30/53] Commits for msvc 16.9

---
 CMakeLists.txt                 |  2 +-
 demo/demo/demo.cpp             |  2 +-
 src/kernels/deformable_conv.cu |  2 +-
 src/utils.cpp                  | 11 ++++++++++-
 4 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index c173eca..77425d3 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -7,7 +7,7 @@ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declara
 endif()
 if(WIN32)
 set(CMAKE_CXX_STANDARD 14)
-set(CMAKE_CXX_FLAGS "/O2 ")
+set(CMAKE_CXX_FLAGS "/O2 /FS ")
 set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
 endif(WIN32)
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include/tkDNN)
diff --git a/demo/demo/demo.cpp b/demo/demo/demo.cpp
index 609affe..c622f2b 100644
--- a/demo/demo/demo.cpp
+++ b/demo/demo/demo.cpp
@@ -22,7 +22,7 @@ int main(int argc, char *argv[]) {
     signal(SIGINT, sig_handler);
 
 
-    std::string net = "yolo3_berkeley.rt";
+    std::string net = "yolo4tiny_fp32.rt";
     if(argc > 1)
         net = argv[1]; 
     std::string input = "../demo/yolo_test.mp4";
diff --git a/src/kernels/deformable_conv.cu b/src/kernels/deformable_conv.cu
index 592c538..4dbc552 100644
--- a/src/kernels/deformable_conv.cu
+++ b/src/kernels/deformable_conv.cu
@@ -18,7 +18,7 @@ inline int GET_BLOCKS(const int N)
 }
 
 
-__device__ float dmcn_im2col_bilinear(const float *bottom_data, const int data_width,
+__device__ __host__  float dmcn_im2col_bilinear(const float *bottom_data, const int data_width,
   const int height, const int width, float h, float w) {
 int h_low = floor(h);
 int w_low = floor(w);
diff --git a/src/utils.cpp b/src/utils.cpp
index bbdf516..52775df 100644
--- a/src/utils.cpp
+++ b/src/utils.cpp
@@ -23,14 +23,23 @@ bool fileExist(const char *fname) {
 void downloadWeightsifDoNotExist(const std::string& input_bin, const std::string& test_folder, const std::string& weights_url){
     if(!fileExist(input_bin.c_str())){
         std::string mkdir_cmd = "mkdir " + test_folder; 
-        std::string wget_cmd = "wget " + weights_url + " -O " + test_folder + "/weights.zip"; 
+        std::string wget_cmd = "curl " + weights_url + " --output " + test_folder + "/weights.zip";
+#ifdef __linux__
         std::string unzip_cmd = "unzip " + test_folder + "/weights.zip -d" + test_folder;
         std::string rm_cmd = "rm " + test_folder + "/weights.zip";
+
+#elif _WIN32
+
+        std::string unzip_cmd = "7z x " + test_folder + "/weights.zip -o" + test_folder;
+#endif 
         int err = 0;
         err = system(mkdir_cmd.c_str());
         err = system(wget_cmd.c_str());
         err = system(unzip_cmd.c_str());
+#ifdef __linux__
         err = system(rm_cmd.c_str());
+#endif
+
     }
 }
 

From 304ab49897938beb3ea6619fb5720bc1aae1280d Mon Sep 17 00:00:00 2001
From: Harshvardhan Chandirasekar <perseusdg@pop-os.localdomain>
Date: Wed, 17 Mar 2021 22:26:26 +0530
Subject: [PATCH 31/53] shared_ptr migrations

---
 CMakeLists.txt            |  2 +-
 include/tkDNN/NetworkRT.h |  7 ++++++-
 include/tkDNN/utils.h     | 11 +++++++++++
 src/NetworkRT.cpp         | 11 +++++++----
 4 files changed, 25 insertions(+), 6 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 77425d3..f478cab 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -3,7 +3,7 @@ cmake_minimum_required(VERSION 3.5)
 project (tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 if(UNIX)
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
 endif()
 if(WIN32)
 set(CMAKE_CXX_STANDARD 14)
diff --git a/include/tkDNN/NetworkRT.h b/include/tkDNN/NetworkRT.h
index 4c6c816..ca012fb 100644
--- a/include/tkDNN/NetworkRT.h
+++ b/include/tkDNN/NetworkRT.h
@@ -6,6 +6,7 @@
 #include "Network.h"
 #include "Layer.h"
 #include "NvInfer.h"
+#include <memory>
 
 namespace tk { namespace dnn {
 
@@ -59,7 +60,8 @@ public:
 #if NV_TENSORRT_MAJOR >= 6  
     nvinfer1::IBuilderConfig *configRT;
 #endif
-    nvinfer1::ICudaEngine *engineRT;
+    std::shared_ptr<nvinfer1::ICudaEngine> engineRT;
+   //nvinfer1::ICudaEngine *engineRT;
     nvinfer1::IExecutionContext *contextRT;
 
     const static int MAX_BUFFERS_RT = 10;
@@ -114,6 +116,9 @@ public:
 
     bool serialize(const char *filename);
     bool deserialize(const char *filename);
+
+
+
 };
 
 }}
diff --git a/include/tkDNN/utils.h b/include/tkDNN/utils.h
index 1404509..edc770e 100644
--- a/include/tkDNN/utils.h
+++ b/include/tkDNN/utils.h
@@ -110,6 +110,17 @@ double t_ns = time_ms.count();
       FatalError(_error.str());                                        \
     }                                                                  \
 }
+struct InferDeleter
+{
+    template <typename T>
+    void operator()(T* obj) const
+    {
+        if (obj)
+        {
+            obj->destroy();
+        }
+    }
+};
 
 typedef enum {
   ERROR_CUDNN = 2,
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index 501ade4..4006caf 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -137,9 +137,11 @@ NetworkRT::NetworkRT(Network *net, const char *name) {
         printCudaMemUsage();
         std::cout<<"Building tensorRT cuda engine...\n";
 #if NV_TENSORRT_MAJOR >= 6                
-        engineRT = builderRT->buildEngineWithConfig(*networkRT, *configRT);
+        //engineRT = builderRT->buildEngineWithConfig(*networkRT, *configRT);
+        engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildEngineWithConfig(*networkRT,*configRT),InferDeleter());
 #else 
-        engineRT = builderRT->buildCudaEngine(*networkRT);
+        //engineRT = builderRT->buildCudaEngine(*networkRT);
+        engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildCudaEngine(*networkRT));
 #endif
         if(engineRT == nullptr)
             FatalError("cloud not build cuda engine")
@@ -561,7 +563,7 @@ ILayer* NetworkRT::convert_layer(ITensor *input, DeformConv2d *l) {
     IPluginLayer *lRT = networkRT->addPlugin(inputs, 2, *plugin);
     checkNULL(lRT);
     lRT->setName( ("Deformable" + std::to_string(l->id)).c_str() );
-    delete(inputs);
+    delete[](inputs);
     // batchnorm
     void *bias_b, *power_b, *mean_b, *variance_b, *scales_b;
     if(dtRT == DataType::kHALF) {
@@ -629,7 +631,8 @@ bool NetworkRT::deserialize(const char *filename) {
 
     pluginFactory = new PluginFactory();
     runtimeRT = createInferRuntime(loggerRT);
-    engineRT = runtimeRT->deserializeCudaEngine(gieModelStream, size, (IPluginFactory *) pluginFactory);
+    //engineRT = runtimeRT->deserializeCudaEngine(gieModelStream, size, (IPluginFactory *) pluginFactory);
+    engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(runtimeRT->deserializeCudaEngine(gieModelStream,size,(IPluginFactory*)pluginFactory),InferDeleter());
     //if (gieModelStream) delete [] gieModelStream;
 
     return true;

From 06787a931fd8897ce8400d62c60393b8ef5fdc85 Mon Sep 17 00:00:00 2001
From: Harshvardhan Chandirasekar <perseusdg@pop-os.localdomain>
Date: Wed, 17 Mar 2021 23:06:50 +0530
Subject: [PATCH 32/53] minor migrations

---
 include/tkDNN/NetworkRT.h | 6 ++++--
 src/NetworkRT.cpp         | 6 ++++--
 2 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/include/tkDNN/NetworkRT.h b/include/tkDNN/NetworkRT.h
index ca012fb..a50cd8c 100644
--- a/include/tkDNN/NetworkRT.h
+++ b/include/tkDNN/NetworkRT.h
@@ -54,8 +54,10 @@ class NetworkRT {
 
 public:
     nvinfer1::DataType dtRT;
-    nvinfer1::IBuilder *builderRT;
-    nvinfer1::IRuntime *runtimeRT;
+    //nvinfer1::IBuilder *builderRT;
+    std::unique_ptr<nvinfer1::IBuilder,InferDeleter> builderRT;
+    //nvinfer1::IRuntime *runtimeRT;
+    std::unique_ptr<nvinfer1::IRuntime,InferDeleter> runtimeRT;
     nvinfer1::INetworkDefinition *networkRT; 
 #if NV_TENSORRT_MAJOR >= 6  
     nvinfer1::IBuilderConfig *configRT;
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index 4006caf..e6eb778 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -33,7 +33,8 @@ NetworkRT::NetworkRT(Network *net, const char *name) {
                    float(NV_TENSORRT_PATCH)/100;
     std::cout<<"New NetworkRT (TensorRT v"<<rt_ver<<")\n";
   
-    builderRT = createInferBuilder(loggerRT);
+   builderRT = std::unique_ptr<nvinfer1::IBuilder,InferDeleter>(createInferBuilder(loggerRT));
+    //builderRT = createInferBuilder(loggerRT);
     std::cout<<"Float16 support: "<<builderRT->platformHasFastFp16()<<"\n";
     std::cout<<"Int8 support: "<<builderRT->platformHasFastInt8()<<"\n";
 #if NV_TENSORRT_MAJOR >= 5
@@ -630,7 +631,8 @@ bool NetworkRT::deserialize(const char *filename) {
     }
 
     pluginFactory = new PluginFactory();
-    runtimeRT = createInferRuntime(loggerRT);
+    //runtimeRT = createInferRuntime(loggerRT);
+    runtimeRT = std::unique_ptr<nvinfer1::IRuntime,InferDeleter>(createInferRuntime(loggerRT));
     //engineRT = runtimeRT->deserializeCudaEngine(gieModelStream, size, (IPluginFactory *) pluginFactory);
     engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(runtimeRT->deserializeCudaEngine(gieModelStream,size,(IPluginFactory*)pluginFactory),InferDeleter());
     //if (gieModelStream) delete [] gieModelStream;

From e94e1f7622d5bc4479036b39e012213acf9c1fdb Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Wed, 24 Mar 2021 22:26:46 +0530
Subject: [PATCH 33/53] tkdnn first patch for windows

---
 .gitignore                                    |   1 +
 demo/demo/demo.cpp                            |   2 +-
 include/tkDNN/DetectionNN.h                   |   1 -
 include/tkDNN/ImuOdom.h                       |   2 +
 include/tkDNN/pluginsRT/ActivationLeakyRT.h   |   3 +-
 include/tkDNN/pluginsRT/ActivationMishRT.h    |   3 +-
 .../tkDNN/pluginsRT/ActivationReLUCeilingRT.h |   3 +-
 include/tkDNN/pluginsRT/ActivationSigmoidRT.h |   3 +-
 include/tkDNN/pluginsRT/DeformableConvRT.h    |   3 +-
 include/tkDNN/pluginsRT/FlattenConcatRT.h     |   3 +-
 .../tkDNN/pluginsRT/MaxPoolingFixedSizeRT.h   |   3 +-
 include/tkDNN/pluginsRT/RegionRT.h            |   3 +-
 include/tkDNN/pluginsRT/ReorgRT.h             |   3 +-
 include/tkDNN/pluginsRT/ReshapeRT.h           |   3 +-
 include/tkDNN/pluginsRT/ResizeLayerRT.h       |   3 +-
 include/tkDNN/pluginsRT/RouteRT.h             |   3 +-
 include/tkDNN/pluginsRT/ShortcutRT.h          |   3 +-
 include/tkDNN/pluginsRT/UpsampleRT.h          |   5 +-
 include/tkDNN/pluginsRT/YoloRT.h              |  36 +++---
 src/NetworkRT.cpp                             | 119 ++++++++++++++----
 20 files changed, 151 insertions(+), 54 deletions(-)

diff --git a/.gitignore b/.gitignore
index 5be5a73..7e5c4ca 100644
--- a/.gitignore
+++ b/.gitignore
@@ -15,3 +15,4 @@ build/
 cmake-build-release/
 demo/COCO_val2017
 demo/BDD100K_val
+/.vs
diff --git a/demo/demo/demo.cpp b/demo/demo/demo.cpp
index c622f2b..f97ead9 100644
--- a/demo/demo/demo.cpp
+++ b/demo/demo/demo.cpp
@@ -131,7 +131,7 @@ int main(int argc, char *argv[]) {
     double mean = 0; 
     
     std::cout<<COL_GREENB<<"\n\nTime stats:\n";
-    std::cout<<"Min: "<<*std::min_element(detNN->stats.begin(), detNN->stats.end())/n_batch<<" ms\n";    
+   std::cout<<"Min: "<<*std::min_element(detNN->stats.begin(), detNN->stats.end())/n_batch<<" ms\n";    
     std::cout<<"Max: "<<*std::max_element(detNN->stats.begin(), detNN->stats.end())/n_batch<<" ms\n";    
     for(int i=0; i<detNN->stats.size(); i++) mean += detNN->stats[i]; mean /= detNN->stats.size();
     std::cout<<"Avg: "<<mean/n_batch<<" ms\t"<<1000/(mean/n_batch)<<" FPS\n"<<COL_END;   
diff --git a/include/tkDNN/DetectionNN.h b/include/tkDNN/DetectionNN.h
index 9ce33c1..fb29be1 100644
--- a/include/tkDNN/DetectionNN.h
+++ b/include/tkDNN/DetectionNN.h
@@ -155,7 +155,6 @@ class DetectionNN {
             int x0, w, x1, y0, h, y1;
             int objClass;
             std::string det_class;
-
             int baseline = 0;
             float font_scale = 0.5;
             int thickness = 2;   
diff --git a/include/tkDNN/ImuOdom.h b/include/tkDNN/ImuOdom.h
index aace012..fa870f3 100644
--- a/include/tkDNN/ImuOdom.h
+++ b/include/tkDNN/ImuOdom.h
@@ -5,6 +5,8 @@
 #ifdef __linux__
 #include <unistd.h>
 #elif _WIN32
+#define _USE_MATH_DEFINES
+#include <math.h>
 #include <Windows.h>
 #endif 
 
diff --git a/include/tkDNN/pluginsRT/ActivationLeakyRT.h b/include/tkDNN/pluginsRT/ActivationLeakyRT.h
index d3f66fb..9e26b2b 100644
--- a/include/tkDNN/pluginsRT/ActivationLeakyRT.h
+++ b/include/tkDNN/pluginsRT/ActivationLeakyRT.h
@@ -52,8 +52,9 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, size);
+		assert(buf == a + getSerializationSize());
 	}
 
 	int size;
diff --git a/include/tkDNN/pluginsRT/ActivationMishRT.h b/include/tkDNN/pluginsRT/ActivationMishRT.h
index 1744ab0..5d660af 100644
--- a/include/tkDNN/pluginsRT/ActivationMishRT.h
+++ b/include/tkDNN/pluginsRT/ActivationMishRT.h
@@ -52,8 +52,9 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, size);
+		assert(buf == a + getSerializationSize());
 	}
 
 	int size;
diff --git a/include/tkDNN/pluginsRT/ActivationReLUCeilingRT.h b/include/tkDNN/pluginsRT/ActivationReLUCeilingRT.h
index 286f22e..50ceb81 100644
--- a/include/tkDNN/pluginsRT/ActivationReLUCeilingRT.h
+++ b/include/tkDNN/pluginsRT/ActivationReLUCeilingRT.h
@@ -51,9 +51,10 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, ceiling);
 		tk::dnn::writeBUF(buf, size);
+		assert(buf = a + getSerializationSize());
 		
 	}
 
diff --git a/include/tkDNN/pluginsRT/ActivationSigmoidRT.h b/include/tkDNN/pluginsRT/ActivationSigmoidRT.h
index 1d47136..bcc58c7 100644
--- a/include/tkDNN/pluginsRT/ActivationSigmoidRT.h
+++ b/include/tkDNN/pluginsRT/ActivationSigmoidRT.h
@@ -52,8 +52,9 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, size);
+		assert(buf == a + getSerializationSize());
 	}
 
 	int size;
diff --git a/include/tkDNN/pluginsRT/DeformableConvRT.h b/include/tkDNN/pluginsRT/DeformableConvRT.h
index 225a24e..5cb2bab 100644
--- a/include/tkDNN/pluginsRT/DeformableConvRT.h
+++ b/include/tkDNN/pluginsRT/DeformableConvRT.h
@@ -116,7 +116,7 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, chunk_dim);
 		tk::dnn::writeBUF(buf, kh);
 		tk::dnn::writeBUF(buf, kw);
@@ -163,6 +163,7 @@ public:
         for(int i=0; i<dim_ones; i++)
     		tk::dnn::writeBUF(buf, aus[i]);
 		free(aus);
+		assert(buf == a + getSerializationSize());
 	}
 
 	cublasStatus_t stat; 
diff --git a/include/tkDNN/pluginsRT/FlattenConcatRT.h b/include/tkDNN/pluginsRT/FlattenConcatRT.h
index 4ebce69..51aa1ab 100644
--- a/include/tkDNN/pluginsRT/FlattenConcatRT.h
+++ b/include/tkDNN/pluginsRT/FlattenConcatRT.h
@@ -65,12 +65,13 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a = buf;
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
 		tk::dnn::writeBUF(buf, rows);
 		tk::dnn::writeBUF(buf, cols);
+		assert(buf == a + getSerializationSize());
 	}
 
 	int c, h, w;
diff --git a/include/tkDNN/pluginsRT/MaxPoolingFixedSizeRT.h b/include/tkDNN/pluginsRT/MaxPoolingFixedSizeRT.h
index 911fca2..0899a34 100644
--- a/include/tkDNN/pluginsRT/MaxPoolingFixedSizeRT.h
+++ b/include/tkDNN/pluginsRT/MaxPoolingFixedSizeRT.h
@@ -55,7 +55,7 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 
 		tk::dnn::writeBUF(buf, this->c);
 		tk::dnn::writeBUF(buf, this->h);
@@ -65,6 +65,7 @@ public:
 		tk::dnn::writeBUF(buf, this->stride_W);
 		tk::dnn::writeBUF(buf, this->winSize);
 		tk::dnn::writeBUF(buf, this->padding);
+		assert(buf == a + getSerializationSize());
 	}
 
 	int n, c, h, w;
diff --git a/include/tkDNN/pluginsRT/RegionRT.h b/include/tkDNN/pluginsRT/RegionRT.h
index f0d127e..8487652 100644
--- a/include/tkDNN/pluginsRT/RegionRT.h
+++ b/include/tkDNN/pluginsRT/RegionRT.h
@@ -73,13 +73,14 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, classes);
 		tk::dnn::writeBUF(buf, coords);
 		tk::dnn::writeBUF(buf, num);
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
+		assert(buf == a + getSerializationSize());
 	}
 
 	int c, h, w;
diff --git a/include/tkDNN/pluginsRT/ReorgRT.h b/include/tkDNN/pluginsRT/ReorgRT.h
index ee85718..c1b529a 100644
--- a/include/tkDNN/pluginsRT/ReorgRT.h
+++ b/include/tkDNN/pluginsRT/ReorgRT.h
@@ -52,11 +52,12 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, stride);
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
+		assert(buf == a + getSerializationSize());
 	}
 
 	int c, h, w, stride;
diff --git a/include/tkDNN/pluginsRT/ReshapeRT.h b/include/tkDNN/pluginsRT/ReshapeRT.h
index 97030db..37017c7 100644
--- a/include/tkDNN/pluginsRT/ReshapeRT.h
+++ b/include/tkDNN/pluginsRT/ReshapeRT.h
@@ -50,11 +50,12 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a = buf;
 		tk::dnn::writeBUF(buf, n);
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
+		assert(buf == a + getSerializationSize());
 	}
 
 	int n, c, h, w;
diff --git a/include/tkDNN/pluginsRT/ResizeLayerRT.h b/include/tkDNN/pluginsRT/ResizeLayerRT.h
index ae87dbf..cde52bf 100644
--- a/include/tkDNN/pluginsRT/ResizeLayerRT.h
+++ b/include/tkDNN/pluginsRT/ResizeLayerRT.h
@@ -52,7 +52,7 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 
 		tk::dnn::writeBUF(buf, o_c);
 		tk::dnn::writeBUF(buf, o_h);
@@ -61,6 +61,7 @@ public:
 		tk::dnn::writeBUF(buf, i_c);
 		tk::dnn::writeBUF(buf, i_h);
 		tk::dnn::writeBUF(buf, i_w);
+		assert(buf == a + getSerializationSize());
 	}
 
 	int i_c, i_h, i_w, o_c, o_h, o_w;
diff --git a/include/tkDNN/pluginsRT/RouteRT.h b/include/tkDNN/pluginsRT/RouteRT.h
index 23f30b7..5a8c170 100644
--- a/include/tkDNN/pluginsRT/RouteRT.h
+++ b/include/tkDNN/pluginsRT/RouteRT.h
@@ -75,7 +75,7 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, groups);
 		tk::dnn::writeBUF(buf, group_id);
 		tk::dnn::writeBUF(buf, in);
@@ -85,6 +85,7 @@ public:
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
+		assert(buf == a + getSerializationSize());
 	}
 
 	static const int MAX_INPUTS = 4;
diff --git a/include/tkDNN/pluginsRT/ShortcutRT.h b/include/tkDNN/pluginsRT/ShortcutRT.h
index 3eadd3f..17f050f 100644
--- a/include/tkDNN/pluginsRT/ShortcutRT.h
+++ b/include/tkDNN/pluginsRT/ShortcutRT.h
@@ -59,13 +59,14 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, bc);
 		tk::dnn::writeBUF(buf, bh);
 		tk::dnn::writeBUF(buf, bw);
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
+		assert(buf == a + getSerializationSize());
 		
 	}
 
diff --git a/include/tkDNN/pluginsRT/UpsampleRT.h b/include/tkDNN/pluginsRT/UpsampleRT.h
index 7a62abc..5350b7e 100644
--- a/include/tkDNN/pluginsRT/UpsampleRT.h
+++ b/include/tkDNN/pluginsRT/UpsampleRT.h
@@ -54,11 +54,14 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
 		tk::dnn::writeBUF(buf, stride);
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
+		std::cout << "Upsample Serialization SIze" << getSerializationSize() << std::endl;
+
+		assert(buf == a + getSerializationSize());
 	}
 
 	int c, h, w, stride;
diff --git a/include/tkDNN/pluginsRT/YoloRT.h b/include/tkDNN/pluginsRT/YoloRT.h
index 9af8587..451d99f 100644
--- a/include/tkDNN/pluginsRT/YoloRT.h
+++ b/include/tkDNN/pluginsRT/YoloRT.h
@@ -89,21 +89,25 @@ public:
 	}
 
 	virtual void serialize(void* buffer) override {
-		char *buf = reinterpret_cast<char*>(buffer);
-		tk::dnn::writeBUF(buf, classes);
-		tk::dnn::writeBUF(buf, num);
-		tk::dnn::writeBUF(buf, n_masks);
-		tk::dnn::writeBUF(buf, scaleXY);
-		tk::dnn::writeBUF(buf, nms_thresh);
-		tk::dnn::writeBUF(buf, nms_kind);
-		tk::dnn::writeBUF(buf, new_coords);
-		tk::dnn::writeBUF(buf, c);
-		tk::dnn::writeBUF(buf, h);
-		tk::dnn::writeBUF(buf, w);
-        for(int i=0; i<n_masks; i++)
-    		tk::dnn::writeBUF(buf, mask[i]);
-        for(int i=0; i<n_masks*2*num; i++)
-    		tk::dnn::writeBUF(buf, bias[i]);
+		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
+		tk::dnn::writeBUF(buf, classes); std::cout << "Classes :" << classes << std::endl;
+		tk::dnn::writeBUF(buf, num); std::cout << "Num : " << num << std::endl;
+		tk::dnn::writeBUF(buf, n_masks); std::cout << "N_Masks" << n_masks << std::endl;
+		tk::dnn::writeBUF(buf, scaleXY); std::cout << "ScaleXY :" << scaleXY << std::endl;
+		tk::dnn::writeBUF(buf, nms_thresh); std::cout << "nms_thresh :" << nms_thresh << std::endl;
+		tk::dnn::writeBUF(buf, nms_kind); std::cout << "nms_kind : " << nms_kind << std::endl;
+		tk::dnn::writeBUF(buf, new_coords); std::cout << "new_coords : " << new_coords << std::endl;
+		tk::dnn::writeBUF(buf, c); std::cout << "C : " << c << std::endl;
+		tk::dnn::writeBUF(buf, h); std::cout << "H : " << h << std::endl;
+		tk::dnn::writeBUF(buf, w); std::cout << "C : " << c << std::endl;
+		for (int i = 0; i < n_masks; i++)
+		{
+			tk::dnn::writeBUF(buf, mask[i]); std::cout << "mask[i] : " << mask[i] << std::endl;
+		}
+		for (int i = 0; i < n_masks * 2 * num; i++)
+		{
+			tk::dnn::writeBUF(buf, bias[i]); std::cout << "bias[i] : " << bias[i] << std::endl;
+		}
 
 		// save classes names
 		for(int i=0; i<classes; i++) {
@@ -113,6 +117,8 @@ public:
 				tk::dnn::writeBUF(buf, tmp[j]);
 			}
 		}
+		std::cout << getSerializationSize() << std::endl;
+		assert(buf == a + getSerializationSize());
 	}
 
 	int c, h, w;
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index e6eb778..29df411 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -643,43 +643,56 @@ bool NetworkRT::deserialize(const char *filename) {
 
 
 IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialData, size_t serialLength) {
-    const char * buf = reinterpret_cast<const char*>(serialData);
+    const char * buf = reinterpret_cast<const char*>(serialData),*bufCheck = buf;
 
     std::string name(layerName);
-    //std::cout<<name<<std::endl;
+    std::cout<<name<<std::endl;
 
     if(name.find("ActivationLeaky") == 0) {
         ActivationLeakyRT *a = new ActivationLeakyRT();
         a->size = readBUF<int>(buf);
+        assert(buf == bufCheck + serialLength);
         return a;
     }
     if(name.find("ActivationMish") == 0) {
         ActivationMishRT *a = new ActivationMishRT();
         a->size = readBUF<int>(buf);
+        assert(buf == bufCheck + serialLength);
         return a;
     }
     if(name.find("ActivationCReLU") == 0) {
-        ActivationReLUCeiling *a = new ActivationReLUCeiling(readBUF<float>(buf));
+        float activationReluTemp = readBUF<float>(buf);
+        //ActivationReLUCeiling *a = new ActivationReLUCeiling(readBUF<float>(buf));
+        ActivationReLUCeiling* a = new ActivationReLUCeiling(activationReluTemp);
         a->size = readBUF<int>(buf);
+        assert(buf == bufCheck + serialLength);
         return a;
     }
 
     if(name.find("Region") == 0) {
-        RegionRT *r = new RegionRT(readBUF<int>(buf),    //classes
+        int classesTemp = readBUF<int>(buf);
+        int coordsTemp = readBUF<int>(buf);
+        int numTemp = readBUF<int>(buf);
+        /*RegionRT *r = new RegionRT(readBUF<int>(buf),    //classes
                                     readBUF<int>(buf),    //coords
-                                    readBUF<int>(buf));   //num
+                                    readBUF<int>(buf));   //num8*/
+        RegionRT* r = new RegionRT(classesTemp, coordsTemp, numTemp);
 
         r->c = readBUF<int>(buf);
         r->h = readBUF<int>(buf);
         r->w = readBUF<int>(buf);
+        assert(buf == bufCheck + serialLength);
         return r;
     } 
 
     if(name.find("Reorg") == 0) {
-        ReorgRT *r = new ReorgRT(readBUF<int>(buf)); //stride
+        int strideTemp = readBUF<int>(buf);
+        //ReorgRT *r = new ReorgRT(readBUF<int>(buf)); //stride
+        ReorgRT *r = new ReorgRT(strideTemp);
         r->c = readBUF<int>(buf);
         r->h = readBUF<int>(buf);
         r->w = readBUF<int>(buf);
+        assert(buf == bufCheck + serialLength);
         return r;
     } 
 
@@ -695,27 +708,46 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         r->h = readBUF<int>(buf);
         r->w = readBUF<int>(buf);
         return r;
+        assert(buf == bufCheck + serialLength);
     } 
 
     if(name.find("Pooling") == 0) {
-        MaxPoolFixedSizeRT *r = new MaxPoolFixedSizeRT( readBUF<int>(buf), //c
+       /* MaxPoolFixedSizeRT *r = new MaxPoolFixedSizeRT( readBUF<int>(buf), //c
                                                         readBUF<int>(buf), //h
                                                         readBUF<int>(buf), //w
                                                         readBUF<int>(buf), //n
                                                         readBUF<int>(buf), //strideH
                                                         readBUF<int>(buf), //strideW
                                                         readBUF<int>(buf), //winSize
-                                                        readBUF<int>(buf)); //padding
+                                                        readBUF<int>(buf)); //padding*/
+
+        int cTemp = readBUF<int>(buf);
+        int hTemp = readBUF<int>(buf);
+        int wTemp = readBUF<int>(buf);
+        int nTemp = readBUF<int>(buf);
+        int strideHTemp = readBUF<int>(buf);
+        int strideWTemp = readBUF<int>(buf);
+        int winSizeTemp = readBUF<int>(buf);
+        int paddingTemp = readBUF<int>(buf);
+
+        MaxPoolFixedSizeRT* r = new MaxPoolFixedSizeRT(cTemp, hTemp, wTemp, nTemp, strideHTemp, strideWTemp, winSizeTemp, paddingTemp);
+        assert(buf == bufCheck + serialLength);
         return r;
     }
 
     if(name.find("Resize") == 0) {
-        ResizeLayerRT *r = new ResizeLayerRT(readBUF<int>(buf), //o_c
+        /*ResizeLayerRT *r = new ResizeLayerRT(readBUF<int>(buf), //o_c
                                             readBUF<int>(buf), //o_h
-                                            readBUF<int>(buf)); //o_w
+                                            readBUF<int>(buf)); //o_w*/
+        int o_cTemp = readBUF<int>(buf);
+        int o_hTemp = readBUF<int>(buf);
+        int o_wTemp = readBUF<int>(buf);
+        ResizeLayerRT* r = new ResizeLayerRT(o_cTemp, o_hTemp, o_wTemp);
+
         r->i_c = readBUF<int>(buf);
         r->i_h = readBUF<int>(buf);
         r->i_w = readBUF<int>(buf);
+        assert(buf == bufCheck + serialLength);
         return r;
     } 
 
@@ -726,6 +758,7 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         r->w = readBUF<int>(buf);
         r->rows = readBUF<int>(buf);
         r->cols = readBUF<int>(buf);
+        assert(buf == bufCheck + serialLength);
         return r;
     } 
 
@@ -737,20 +770,33 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         new_dim.h = readBUF<int>(buf);
         new_dim.w = readBUF<int>(buf);
         ReshapeRT *r = new ReshapeRT(new_dim); 
+        assert(buf == bufCheck + serialLength);
         
         return r;
     } 
 
     if(name.find("Yolo") == 0) {
-        YoloRT *r = new YoloRT(readBUF<int>(buf),    //classes
-                                readBUF<int>(buf),   //num
-                                nullptr, //yolo
-                                readBUF<int>(buf), //n_masks
-                                readBUF<float>(buf), //scale_xy
-                                readBUF<float>(buf),  //nms_thresh 
-                                readBUF<int>(buf),  //nms_kind
-                                readBUF<int>(buf)  //new_coords
-                                );   
+
+        int classes_temp = readBUF<int>(buf);
+        int num_temp = readBUF<int>(buf);
+        int n_masks_temp = readBUF<int>(buf);
+        float scale_xy_temp = readBUF<float>(buf);
+        float nms_thresh_temp = readBUF<float>(buf);
+        int nms_kind_temp = readBUF<int>(buf);
+        int new_coords_temp = readBUF<int>(buf);
+        std::cout << classes_temp << ":" << num_temp << ":" << ":" << n_masks_temp << ":" << nms_thresh_temp << ":" << nms_kind_temp << ":" << new_coords_temp << std::endl;
+
+       YoloRT *r = new YoloRT(classes_temp,num_temp,nullptr,n_masks_temp,scale_xy_temp,nms_thresh_temp,nms_kind_temp,new_coords_temp);  
+
+       /* std::cout << "classes : " << r->classes;
+        std::cout << "num : " << r->num;
+        std::cout << "n_masks : " << r->n_masks;
+        std::cout << "scalexy : " << r->scaleXY;
+        std::cout << "nms_thresh : " << r->nms_thresh;
+        std::cout << "nms_kind : " << r->nms_kind;
+        std::cout << "new_coords : " << r->new_coords;*/
+
+
         r->c = readBUF<int>(buf);
         r->h = readBUF<int>(buf);
         r->w = readBUF<int>(buf);
@@ -767,36 +813,62 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
 				tmp[j] = readBUF<char>(buf);
             r->classesNames[i] = std::string(tmp);
 		}
+        assert(buf == bufCheck + serialLength);
 
         yolos[n_yolos++] = r;
         return r;
     } 
     if(name.find("Upsample") == 0) {
-        UpsampleRT *r = new UpsampleRT(readBUF<int>(buf)); //stride
+        //UpsampleRT *r = new UpsampleRT(readBUF<int>(buf)); //stride
+        int strideTemp = readBUF<int>(buf);
+        UpsampleRT* r = new UpsampleRT(strideTemp);
         r->c = readBUF<int>(buf);
         r->h = readBUF<int>(buf);
         r->w = readBUF<int>(buf);
+        assert(buf == bufCheck + serialLength);
         return r;
     } 
 
     if(name.find("Route") == 0) {
-        RouteRT *r = new RouteRT(readBUF<int>(buf),readBUF<int>(buf));
+        //RouteRT *r = new RouteRT(readBUF<int>(buf),readBUF<int>(buf));
+        int groupsTemp = readBUF<int>(buf);
+        int group_idTemp = readBUF<int>(buf);
+        RouteRT* r = new RouteRT(groupsTemp, group_idTemp);
         r->in = readBUF<int>(buf);
         for(int i=0; i<RouteRT::MAX_INPUTS; i++)
             r->c_in[i] = readBUF<int>(buf);
         r->c = readBUF<int>(buf);
         r->h = readBUF<int>(buf);
         r->w = readBUF<int>(buf);
+        assert(buf == bufCheck + serialLength);
         return r;
     } 
 
     if(name.find("Deformable") == 0) {
-        DeformableConvRT *r = new DeformableConvRT(readBUF<int>(buf), readBUF<int>(buf), readBUF<int>(buf),
+        /*DeformableConvRT *r = new DeformableConvRT(readBUF<int>(buf), readBUF<int>(buf), readBUF<int>(buf),
                                                     readBUF<int>(buf), readBUF<int>(buf), readBUF<int>(buf),
                                                     readBUF<int>(buf), readBUF<int>(buf),
                                                     readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),
                                                     readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),
-                                                    nullptr); 
+                                                    nullptr); */
+        int chuck_dimTemp = readBUF<int>(buf);
+        int khTemp = readBUF<int>(buf);
+        int kwTemp = readBUF<int>(buf);
+        int shTemp = readBUF<int>(buf);
+        int swTemp = readBUF<int>(buf);
+        int phTemp = readBUF<int>(buf);
+        int pwTemp = readBUF<int>(buf);
+        int deformableGroupTemp = readBUF<int>(buf);
+        int i_nTemp = readBUF<int>(buf);
+        int i_cTemp = readBUF<int>(buf);
+        int i_hTemp = readBUF<int>(buf);
+        int i_wTemp = readBUF<int>(buf);
+        int o_nTemp = readBUF<int>(buf);
+        int o_cTemp = readBUF<int>(buf);
+        int o_hTemp = readBUF<int>(buf);
+        int o_wTemp = readBUF<int>(buf);
+
+        DeformableConvRT* r = new DeformableConvRT(chuck_dimTemp, khTemp, kwTemp, shTemp, swTemp, phTemp, pwTemp, deformableGroupTemp, i_nTemp, i_cTemp, i_hTemp, i_wTemp, o_nTemp, o_cTemp, o_hTemp, o_wTemp, nullptr);
         dnnType *aus = new dnnType[r->chunk_dim*2];
         for(int i=0; i<r->chunk_dim*2; i++)
     		aus[i] = readBUF<dnnType>(buf);
@@ -827,6 +899,7 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
     		aus[i] = readBUF<dnnType>(buf);
 		checkCuda( cudaMemcpy(r->ones_d2, aus, sizeof(dnnType)*r->dim_ones, cudaMemcpyHostToDevice) );
         free(aus);
+        assert(buf == bufCheck + serialLength);
         return r;
     } 
 

From 78859fe19109f265489090911ee57629bf31ff67 Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Thu, 25 Mar 2021 16:38:53 +0530
Subject: [PATCH 34/53] timer fix

---
 .gitignore            | 4 ++++
 CMakeLists.txt        | 2 +-
 include/tkDNN/utils.h | 2 +-
 3 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/.gitignore b/.gitignore
index 7e5c4ca..c1d362c 100644
--- a/.gitignore
+++ b/.gitignore
@@ -16,3 +16,7 @@ cmake-build-release/
 demo/COCO_val2017
 demo/BDD100K_val
 /.vs
+cmake-build-minsizerel/*
+scripts/COCO_val2017/*
+scripts/COCO_val2017.zip
+scripts/all_labels.txt
\ No newline at end of file
diff --git a/CMakeLists.txt b/CMakeLists.txt
index f478cab..579a8bc 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -7,7 +7,7 @@ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -fPIC -Wno-deprecated-declara
 endif()
 if(WIN32)
 set(CMAKE_CXX_STANDARD 14)
-set(CMAKE_CXX_FLAGS "/O2 /FS ")
+set(CMAKE_CXX_FLAGS "/O1 /FS")
 set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
 endif(WIN32)
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include/tkDNN)
diff --git a/include/tkDNN/utils.h b/include/tkDNN/utils.h
index edc770e..ce4a617 100644
--- a/include/tkDNN/utils.h
+++ b/include/tkDNN/utils.h
@@ -60,7 +60,7 @@
 #define TKDNN_TSTART auto start = std::chrono::high_resolution_clock::now();
 #define TKDNN_TSTOP auto stop = std::chrono::high_resolution_clock::now(); \
 std::chrono::duration<double> duration = stop -start;                      \
-auto time_ms = std::chrono::duration_cast<std::chrono::microseconds>(duration);\
+auto time_ms = std::chrono::duration_cast<std::chrono::milliseconds>(duration);\
 double t_ns = time_ms.count();
 #endif
 

From 44b71ae6f33acbaf09fd8a4064d15b8395768cfb Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Thu, 25 Mar 2021 20:44:43 +0530
Subject: [PATCH 35/53] ReadMe.md windows changes

---
 CMakeLists.txt |  2 +-
 README.md      | 85 ++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 86 insertions(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 579a8bc..1b7ed63 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -7,7 +7,7 @@ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -fPIC -Wno-deprecated-declara
 endif()
 if(WIN32)
 set(CMAKE_CXX_STANDARD 14)
-set(CMAKE_CXX_FLAGS "/O1 /FS")
+set(CMAKE_CXX_FLAGS "/O1 /FS /EHsc")
 set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
 endif(WIN32)
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include/tkDNN)
diff --git a/README.md b/README.md
index 84e0037..242220a 100644
--- a/README.md
+++ b/README.md
@@ -80,6 +80,13 @@ Results for COCO val 2017 (5k images), on RTX 2080Ti, with conf threshold=0.001
   - [mAP demo](#map-demo)
   - [Existing tests and supported networks](#existing-tests-and-supported-networks)
   - [References](#references)
+  - [tkDNN on Windows 10 (experimental)](#tkdnn-on-windows)
+    - [Dependencies](#dependencies)
+    - [Compiling tkDNN on Windows](#tkdnn-windows-compile)
+    - [Run the demo on Windows](#run-the-demo-on-windows)
+    - [FP16 interference windows](#fp16-windows)
+    - [INT8 interference windows](#int8-windows)
+  
 
 
 
@@ -355,6 +362,84 @@ This demo also creates a json file named ```net_name_COCO_res.json``` containing
 | yolo4tiny             | Yolov4 tiny <sup>9</sup>                           | [COCO 2017](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/iRnc4pSqmx78gJs/download)     |
 | yolo4x             | Yolov4x-mish  <sup>9</sup>                          | [COCO 2017](http://cocodataset.org/)                          | 80        | 672x672       | [weights](https://cloud.hipert.unimore.it/s/BLPpiAigZJLorQD/download)     |
 
+##tkDNN on Windows 10 (experimental)
+
+### Dependencies
+This branch should work on every NVIDIA GPU supported in windows with the following dependencies:
+
+* WINDOWS 10 1803 or HIGHER 
+* CUDA 10.0 (Recommended CUDA 11.0 +)
+* CUDNN 7.6 (Recommended CUDNN 8.0.0 +)
+* TENSORRT 6.0.1 (Recommended TENSORRT 7.1 +)
+* OPENCV 3.4 (Recommended OPENCV 4.2.0 +)
+* MSVC 16.7 (Recommended MSVC 16.8/16.9)
+* YAML-CPP 0.5.2 
+* EIGEN3
+* 7ZIP (ADD TO PATH)
+* NINJA 1.10
+
+All the above mentioned dependencies except 7ZIP can be installed using Microsoft's [VCPKG](https://github.com/microsoft/vcpkg.git) .
+After bootstrapping VCPKG the dependencies can be built and installed using the following command :
+
+```vcpkg.exe install opencv4[tbb,jpeg,tiff,opengl,openmp,png,ffmpeg]:x64-windows yaml-cpp:x64-windows eigen3:x64-windows --x-install-root=C:\opt --x-buildtrees-root=C:\temp_vcpkg_build```
+
+After VCPKG finishes building and installing all the packages delete C:\temp_vcpkg_build and add C:\opt\x64-windows\bin and C:\opt\x64-windows\debug\bin to path 
+
+### Compiling tkDNN on Windows 
+
+tkDNN is built with cmake(3.15+) on windows along with ninja.Msbuild and NMake Makefiles are drastically slower when compiling the library compared to windows
+```
+git clone https://git.hipert.unimore.it/research-cv-chandirasekar/tkdnn-windows.git
+cd tkdnn-windows
+mkdir build
+cd build
+cmake -DCMAKE_BUILD_TYPE=Release -G"Ninja" ..
+ninja -j4
+```
+
+### Run the demo on Windows 
+
+This example uses yolo4_tiny.\
+To run the object detection file create .rt file bu running:
+```
+.\test_yolo4tiny.exe
+```
+
+Once the rt file has been successfully create,run the demo using the following command:
+```
+.\demo.exe yolo4tiny_fp32.rt ..\demo\yolo_test.mp4 y 
+```
+ For general info on more demo paramters,check Run the demo section on top 
+
+### FP16 interference windows 
+
+This is an untested feature on windows.To run the object detection demo with FP16 interference follow the below steps(example with yolo4tiny):
+```
+set TKDNN_MODE=FP16
+del /f yolo4tiny_fp16.rt
+.\test_yolo4tiny.exe
+.\demo.exe yolo4tiny_fp16.rt ..\demo\yolo_test.mp4
+```
+
+### INT8 interference windows 
+To run object detection demo with INT8 (example with yolo4tiny):
+```
+set TKDNN_MODE=INT8
+set TKDNN_CALIB_LABEL_PATH=..\demo\COCO_val2017\all_labels.txt
+set TKDNN_CALIB_IMG_PATH=..\demo\COCO_val2017\all_images.txt
+del /f  yolo4tiny_int8.rt        # be sure to delete(or move) old tensorRT files
+.\test_yolo4tiny.exe           # run the yolo test (is slow)
+.\demo.exe yolo4tiny_int8.rt ..\demo\yolo_test.mp4 y
+
+```
+
+
+
+
+
+
+
+
 
 ## References
 

From f3d159143025672d5cd1bc91a89370dac1d902c2 Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Thu, 25 Mar 2021 20:46:12 +0530
Subject: [PATCH 36/53] ReadMe.md windows changes

---
 README.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index 242220a..f685b58 100644
--- a/README.md
+++ b/README.md
@@ -81,7 +81,7 @@ Results for COCO val 2017 (5k images), on RTX 2080Ti, with conf threshold=0.001
   - [Existing tests and supported networks](#existing-tests-and-supported-networks)
   - [References](#references)
   - [tkDNN on Windows 10 (experimental)](#tkdnn-on-windows)
-    - [Dependencies](#dependencies)
+    - [Dependencies-Windows](#dependencies-windows)
     - [Compiling tkDNN on Windows](#tkdnn-windows-compile)
     - [Run the demo on Windows](#run-the-demo-on-windows)
     - [FP16 interference windows](#fp16-windows)
@@ -364,7 +364,7 @@ This demo also creates a json file named ```net_name_COCO_res.json``` containing
 
 ##tkDNN on Windows 10 (experimental)
 
-### Dependencies
+### Dependencies-Windows 
 This branch should work on every NVIDIA GPU supported in windows with the following dependencies:
 
 * WINDOWS 10 1803 or HIGHER 

From f12ec3c935ce4fdcb45c966e9b4f753454d630b9 Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Fri, 26 Mar 2021 12:18:46 +0530
Subject: [PATCH 37/53] added yolo4x and yolo4-csp from the github repo and
 download file corrections

---
 include/tkDNN/Layer.h                         |    7 +-
 include/tkDNN/NetworkRT.h                     |    1 +
 .../tkDNN/pluginsRT/ActivationLogisticRT.h    |   60 +
 include/tkDNN/pluginsRT/YoloRT.h              |   29 +-
 src/Activation.cpp                            |    6 +-
 src/DarknetParser.cpp                         |    1 +
 src/NetworkRT.cpp                             |   21 +-
 src/Yolo.cpp                                  |   21 +-
 tests/darknet/cfg/yolo4-csp.cfg               | 1279 +++++++++++++++++
 tests/darknet/cfg/yolo4x.cfg                  |   21 +-
 tests/darknet/yolo4-csp.cpp                   |   36 +
 tests/darknet/yolo4x.cpp                      |    2 +-
 12 files changed, 1444 insertions(+), 40 deletions(-)
 create mode 100644 include/tkDNN/pluginsRT/ActivationLogisticRT.h
 create mode 100644 tests/darknet/cfg/yolo4-csp.cfg
 create mode 100644 tests/darknet/yolo4-csp.cpp

diff --git a/include/tkDNN/Layer.h b/include/tkDNN/Layer.h
index 25c4565..e097372 100644
--- a/include/tkDNN/Layer.h
+++ b/include/tkDNN/Layer.h
@@ -19,6 +19,7 @@ enum layerType_t {
     LAYER_ACTIVATION_CRELU,
     LAYER_ACTIVATION_LEAKY,
     LAYER_ACTIVATION_MISH,
+    LAYER_ACTIVATION_LOGISTIC,
     LAYER_FLATTEN,
     LAYER_RESHAPE,
     LAYER_MULADD,
@@ -68,6 +69,7 @@ public:
             case LAYER_ACTIVATION_CRELU:    return "ActivationCReLU";
             case LAYER_ACTIVATION_LEAKY:    return "ActivationLeaky";
             case LAYER_ACTIVATION_MISH:     return "ActivationMish";
+            case LAYER_ACTIVATION_LOGISTIC: return "ActivationLogistic";
             case LAYER_FLATTEN:             return "Flatten";
             case LAYER_RESHAPE:             return "Reshape";
             case LAYER_MULADD:              return "MulAdd";
@@ -212,7 +214,8 @@ public:
 typedef enum {
     ACTIVATION_ELU     = 100,
     ACTIVATION_LEAKY   = 101,
-    ACTIVATION_MISH   = 102
+    ACTIVATION_MISH   = 102,
+    ACTIVATION_LOGISTIC   = 103
 } tkdnnActivationMode_t;
 
 /**
@@ -233,6 +236,8 @@ public:
             return LAYER_ACTIVATION_LEAKY;
         else if (act_mode == ACTIVATION_MISH)
             return LAYER_ACTIVATION_MISH;
+        else if (act_mode == ACTIVATION_LOGISTIC)
+            return LAYER_ACTIVATION_LOGISTIC;
         else
             return LAYER_ACTIVATION;
          };
diff --git a/include/tkDNN/NetworkRT.h b/include/tkDNN/NetworkRT.h
index a50cd8c..b39360c 100644
--- a/include/tkDNN/NetworkRT.h
+++ b/include/tkDNN/NetworkRT.h
@@ -27,6 +27,7 @@ using namespace nvinfer1;
 #include "pluginsRT/ActivationLeakyRT.h"
 #include "pluginsRT/ActivationReLUCeilingRT.h"
 #include "pluginsRT/ActivationMishRT.h"
+#include "pluginsRT/ActivationLogisticRT.h"
 #include "pluginsRT/ReorgRT.h"
 #include "pluginsRT/RegionRT.h"
 #include "pluginsRT/RouteRT.h"
diff --git a/include/tkDNN/pluginsRT/ActivationLogisticRT.h b/include/tkDNN/pluginsRT/ActivationLogisticRT.h
new file mode 100644
index 0000000..83f62ff
--- /dev/null
+++ b/include/tkDNN/pluginsRT/ActivationLogisticRT.h
@@ -0,0 +1,60 @@
+#include<cassert>
+#include "../kernels.h"
+
+class ActivationLogisticRT : public IPlugin {
+
+public:
+    ActivationLogisticRT() {
+
+
+    }
+
+    ~ActivationLogisticRT(){
+
+    }
+
+    int getNbOutputs() const override {
+        return 1;
+    }
+
+    Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) override {
+        return inputs[0];
+    }
+
+    void configure(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, int maxBatchSize) override {
+        size = 1;
+        for(int i=0; i<outputDims[0].nbDims; i++)
+            size *= outputDims[0].d[i];
+    }
+
+    int initialize() override {
+
+        return 0;
+    }
+
+    virtual void terminate() override {
+    }
+
+    virtual size_t getWorkspaceSize(int maxBatchSize) const override {
+        return 0;
+    }
+
+    virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override {
+
+        activationLOGISTICForward((dnnType*)reinterpret_cast<const dnnType*>(inputs[0]),
+                                  reinterpret_cast<dnnType*>(outputs[0]), batchSize*size, stream);
+        return 0;
+    }
+
+
+    virtual size_t getSerializationSize() override {
+        return 1*sizeof(int);
+    }
+
+    virtual void serialize(void* buffer) override {
+        char *buf = reinterpret_cast<char*>(buffer);
+        tk::dnn::writeBUF(buf, size);
+    }
+
+    int size;
+};
\ No newline at end of file
diff --git a/include/tkDNN/pluginsRT/YoloRT.h b/include/tkDNN/pluginsRT/YoloRT.h
index 451d99f..0dd26e1 100644
--- a/include/tkDNN/pluginsRT/YoloRT.h
+++ b/include/tkDNN/pluginsRT/YoloRT.h
@@ -64,20 +64,23 @@ public:
 
 		checkCuda( cudaMemcpyAsync(dstData, srcData, batchSize*c*h*w*sizeof(dnnType), cudaMemcpyDeviceToDevice, stream));
 
-		for (int b = 0; b < batchSize; ++b){
-			for(int n = 0; n < n_masks; ++n){
-				int index = entry_index(b, n*w*h, 0);
-				if (new_coords == 1)
-					activationLOGISTICForward(srcData + index, dstData + index, 4*w*h, stream); //x,y,w,h
-				else
-					activationLOGISTICForward(srcData + index, dstData + index, 2*w*h, stream); //x,y
 
-				if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
-				
-				index = entry_index(b, n*w*h, 4);
-				activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*w*h, stream);
-			}
-		}
+        for (int b = 0; b < batchSize; ++b){
+            for(int n = 0; n < n_masks; ++n){
+                int index = entry_index(b, n*w*h, 0);
+                if (new_coords == 1){
+                    if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
+                }
+                else{
+                    activationLOGISTICForward(srcData + index, dstData + index, 2*w*h, stream); //x,y
+
+                    if (this->scaleXY != 1) scalAdd(dstData + index, 2 * w*h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
+
+                    index = entry_index(b, n*w*h, 4);
+                    activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*w*h, stream);
+                }
+            }
+        }
 
 		//std::cout<<"YOLO END\n";
 		return 0;
diff --git a/src/Activation.cpp b/src/Activation.cpp
index 28c7624..4219271 100644
--- a/src/Activation.cpp
+++ b/src/Activation.cpp
@@ -52,7 +52,11 @@ dnnType* Activation::infer(dataDim_t &dim, dnnType* srcData) {
     else if(act_mode == ACTIVATION_MISH) {
         activationMishForward(srcData, dstData, dim.tot());
 
-    } else {
+    }
+    else if(act_mode == ACTIVATION_LOGISTIC) {
+        activationLOGISTICForward(srcData, dstData, dim.tot());
+
+    }else {
         dnnType alpha = dnnType(1);
         dnnType beta  = dnnType(0);
         checkCUDNN( cudnnActivationForward(net->cudnnHandle,
diff --git a/src/DarknetParser.cpp b/src/DarknetParser.cpp
index 7b5410c..69b6b29 100644
--- a/src/DarknetParser.cpp
+++ b/src/DarknetParser.cpp
@@ -187,6 +187,7 @@ namespace tk { namespace dnn {
             if(f.activation == "relu") act = tkdnnActivationMode_t(CUDNN_ACTIVATION_RELU);
             else if(f.activation == "leaky") act = tk::dnn::ACTIVATION_LEAKY;
             else if(f.activation == "mish") act = tk::dnn::ACTIVATION_MISH;
+            else if(f.activation == "logistic") act = tk::dnn::ACTIVATION_LOGISTIC;
             else { FatalError("activation not supported: " + f.activation); }
             netLayers[netLayers.size()-1] = new tk::dnn::Activation(net, act);
         };
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index 29df411..1e0b062 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -229,7 +229,7 @@ ILayer* NetworkRT::convert_layer(ITensor *input, Layer *l) {
         return convert_layer(input, (Conv2d*) l);
     if(type == LAYER_POOLING)
         return convert_layer(input, (Pooling*) l);
-    if(type == LAYER_ACTIVATION || type == LAYER_ACTIVATION_CRELU || type == LAYER_ACTIVATION_LEAKY || type == LAYER_ACTIVATION_MISH)
+    if(type == LAYER_ACTIVATION || type == LAYER_ACTIVATION_CRELU || type == LAYER_ACTIVATION_LEAKY || type == LAYER_ACTIVATION_MISH || type == LAYER_ACTIVATION_LOGISTIC)
         return convert_layer(input, (Activation*) l);
     if(type == LAYER_SOFTMAX)
         return convert_layer(input, (Softmax*) l);
@@ -424,6 +424,12 @@ ILayer* NetworkRT::convert_layer(ITensor *input, Activation *l) {
         checkNULL(lRT);
         return lRT;
     }
+    else if(l->act_mode == ACTIVATION_LOGISTIC) {
+        IPlugin *plugin = new ActivationLogisticRT();
+        IPluginLayer *lRT = networkRT->addPlugin(&input, 1, *plugin);
+        checkNULL(lRT);
+        return lRT;
+    }
     else {
         FatalError("this Activation mode is not yet implemented");
         return NULL;
@@ -660,6 +666,11 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         assert(buf == bufCheck + serialLength);
         return a;
     }
+    if(name.find("ActivationLogistic") == 0) {
+        ActivationLogisticRT *a = new ActivationLogisticRT();
+        a->size = readBUF<int>(buf);
+        return a;
+    }
     if(name.find("ActivationCReLU") == 0) {
         float activationReluTemp = readBUF<float>(buf);
         //ActivationReLUCeiling *a = new ActivationReLUCeiling(readBUF<float>(buf));
@@ -784,17 +795,9 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         float nms_thresh_temp = readBUF<float>(buf);
         int nms_kind_temp = readBUF<int>(buf);
         int new_coords_temp = readBUF<int>(buf);
-        std::cout << classes_temp << ":" << num_temp << ":" << ":" << n_masks_temp << ":" << nms_thresh_temp << ":" << nms_kind_temp << ":" << new_coords_temp << std::endl;
 
        YoloRT *r = new YoloRT(classes_temp,num_temp,nullptr,n_masks_temp,scale_xy_temp,nms_thresh_temp,nms_kind_temp,new_coords_temp);  
 
-       /* std::cout << "classes : " << r->classes;
-        std::cout << "num : " << r->num;
-        std::cout << "n_masks : " << r->n_masks;
-        std::cout << "scalexy : " << r->scaleXY;
-        std::cout << "nms_thresh : " << r->nms_thresh;
-        std::cout << "nms_kind : " << r->nms_kind;
-        std::cout << "new_coords : " << r->new_coords;*/
 
 
         r->c = readBUF<int>(buf);
diff --git a/src/Yolo.cpp b/src/Yolo.cpp
index 6d0b546..1eb7843 100644
--- a/src/Yolo.cpp
+++ b/src/Yolo.cpp
@@ -73,8 +73,8 @@ Yolo::box get_yolo_box(float *x, float *biases, int n, int index, int i, int j,
         b.h = exp(x[index + 3*stride]) * biases[2*n+1] / h;
     }
     else{
-        b.x = (i + x[index + 0 * stride] * 2 - 0.5) / lw;
-        b.y = (j + x[index + 1 * stride] * 2 - 0.5) / lh;
+        b.x = (i + x[index + 0 * stride] ) / lw;
+        b.y = (j + x[index + 1 * stride] ) / lh;
         b.w = x[index + 2 * stride] * x[index + 2 * stride] * 4 * biases[2 * n] / w;
         b.h = x[index + 3 * stride] * x[index + 3 * stride] * 4 * biases[2 * n + 1] / h;
     }
@@ -88,15 +88,18 @@ dnnType* Yolo::infer(dataDim_t &dim, dnnType* srcData) {
     for (int b = 0; b < dim.n; ++b){
         for(int n = 0; n < n_masks; ++n){
             int index = entry_index(b, n*dim.w*dim.h, 0, classes, input_dim, output_dim);
-            if (new_coords == 1)
-                activationLOGISTICForward(srcData + index, dstData + index, 4*dim.w*dim.h);
-            else
+            std::cout<<"new_coords"<<new_coords<<std::endl;
+            if (new_coords == 1){
+                if (this->scaleXY != 1) scalAdd(dstData + index, 2 * dim.w*dim.h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
+            }
+            else{
                 activationLOGISTICForward(srcData + index, dstData + index, 2*dim.w*dim.h);
 
-            if (this->scaleXY != 1) scalAdd(dstData + index, 2 * dim.w*dim.h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
-            
-            index = entry_index(b, n*dim.w*dim.h, 4, classes, input_dim, output_dim);
-            activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*dim.w*dim.h);
+                if (this->scaleXY != 1) scalAdd(dstData + index, 2 * dim.w*dim.h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
+
+                index = entry_index(b, n*dim.w*dim.h, 4, classes, input_dim, output_dim);
+                activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*dim.w*dim.h);
+            }
         }
     }
 
diff --git a/tests/darknet/cfg/yolo4-csp.cfg b/tests/darknet/cfg/yolo4-csp.cfg
new file mode 100644
index 0000000..887898e
--- /dev/null
+++ b/tests/darknet/cfg/yolo4-csp.cfg
@@ -0,0 +1,1279 @@
+[net]
+# Testing
+#batch=1
+#subdivisions=1
+# Training
+batch=64
+subdivisions=8
+width=512
+height=512
+channels=3
+momentum=0.949
+decay=0.0005
+angle=0
+saturation = 1.5
+exposure = 1.5
+hue=.1
+
+learning_rate=0.001
+burn_in=1000
+max_batches = 500500
+policy=steps
+steps=400000,450000
+scales=.1,.1
+
+mosaic=1
+
+letter_box=1
+
+ema_alpha=0.9998
+
+#optimized_memory=1
+
+#23:104x104 54:52x52 85:26x26 104:13x13 for 416
+
+
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=3
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=2
+pad=1
+activation=mish
+
+#[convolutional]
+#batch_normalize=1
+#filters=64
+#size=1
+#stride=1
+#pad=1
+#activation=mish
+
+#[route]
+#layers = -2
+
+#[convolutional]
+#batch_normalize=1
+#filters=64
+#size=1
+#stride=1
+#pad=1
+#activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=32
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+#[convolutional]
+#batch_normalize=1
+#filters=64
+#size=1
+#stride=1
+#pad=1
+#activation=mish
+
+#[route]
+#layers = -1,-7
+
+#[convolutional]
+#batch_normalize=1
+#filters=64
+#size=1
+#stride=1
+#pad=1
+#activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=64
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-10
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-28
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-28
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+# Downsample
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=3
+stride=2
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=3
+stride=1
+pad=1
+activation=mish
+
+[shortcut]
+from=-3
+activation=linear
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1,-16
+
+[convolutional]
+batch_normalize=1
+filters=1024
+size=1
+stride=1
+pad=1
+activation=mish
+
+##########################
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+### SPP ###
+[maxpool]
+stride=1
+size=5
+
+[route]
+layers=-2
+
+[maxpool]
+stride=1
+size=9
+
+[route]
+layers=-4
+
+[maxpool]
+stride=1
+size=13
+
+[route]
+layers=-1,-3,-5,-6
+### End SPP ###
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[route]
+layers = -1, -13
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[upsample]
+stride=2
+
+[route]
+layers = 79
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1, -3
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[route]
+layers = -1, -6
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[upsample]
+stride=2
+
+[route]
+layers = 48
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -1, -3
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=128
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=128
+activation=mish
+
+[route]
+layers = -1, -6
+
+[convolutional]
+batch_normalize=1
+filters=128
+size=1
+stride=1
+pad=1
+activation=mish
+
+##########################
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=logistic
+
+
+[yolo]
+mask = 0,1,2
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=80
+num=9
+jitter=.1
+scale_x_y = 2.0
+objectness_smooth=0
+ignore_thresh = .7
+truth_thresh = 1
+#random=1
+resize=1.5
+iou_thresh=0.2
+iou_normalizer=0.05
+cls_normalizer=0.5
+obj_normalizer=4.0
+iou_loss=ciou
+nms_kind=diounms
+beta_nms=0.6
+new_coords=1
+max_delta=5
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=2
+pad=1
+filters=256
+activation=mish
+
+[route]
+layers = -1, -20
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=256
+activation=mish
+
+[route]
+layers = -1,-6
+
+[convolutional]
+batch_normalize=1
+filters=256
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=logistic
+
+
+[yolo]
+mask = 3,4,5
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=80
+num=9
+jitter=.1
+scale_x_y = 2.0
+objectness_smooth=1
+ignore_thresh = .7
+truth_thresh = 1
+#random=1
+resize=1.5
+iou_thresh=0.2
+iou_normalizer=0.05
+cls_normalizer=0.5
+obj_normalizer=1.0
+iou_loss=ciou
+nms_kind=diounms
+beta_nms=0.6
+new_coords=1
+max_delta=5
+
+[route]
+layers = -4
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=2
+pad=1
+filters=512
+activation=mish
+
+[route]
+layers = -1, -49
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[route]
+layers = -2
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=512
+activation=mish
+
+[route]
+layers = -1,-6
+
+[convolutional]
+batch_normalize=1
+filters=512
+size=1
+stride=1
+pad=1
+activation=mish
+
+[convolutional]
+batch_normalize=1
+size=3
+stride=1
+pad=1
+filters=1024
+activation=mish
+
+[convolutional]
+size=1
+stride=1
+pad=1
+filters=255
+activation=logistic
+
+
+[yolo]
+mask = 6,7,8
+anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
+classes=80
+num=9
+jitter=.1
+scale_x_y = 2.0
+objectness_smooth=1
+ignore_thresh = .7
+truth_thresh = 1
+#random=1
+resize=1.5
+iou_thresh=0.2
+iou_normalizer=0.05
+cls_normalizer=0.5
+obj_normalizer=0.4
+iou_loss=ciou
+nms_kind=diounms
+beta_nms=0.6
+new_coords=1
+max_delta=2
\ No newline at end of file
diff --git a/tests/darknet/cfg/yolo4x.cfg b/tests/darknet/cfg/yolo4x.cfg
index 89f2564..f6604f6 100644
--- a/tests/darknet/cfg/yolo4x.cfg
+++ b/tests/darknet/cfg/yolo4x.cfg
@@ -5,8 +5,8 @@
 # Training
 batch=64
 subdivisions=8
-width=672
-height=672
+width=640
+height=640
 channels=3
 momentum=0.949
 decay=0.0005
@@ -15,7 +15,7 @@ saturation = 1.5
 exposure = 1.5
 hue=.1
 
-learning_rate=0.00261
+learning_rate=0.001
 burn_in=1000
 max_batches = 500500
 policy=steps
@@ -26,6 +26,8 @@ mosaic=1
 
 letter_box=1
 
+#optimized_memory=1
+
 [convolutional]
 batch_normalize=1
 filters=32
@@ -1131,6 +1133,7 @@ size=1
 stride=1
 pad=1
 activation=mish
+stopbackward=800
 
 ##########################
 
@@ -1147,7 +1150,7 @@ size=1
 stride=1
 pad=1
 filters=255
-activation=linear
+activation=logistic
 
 
 [yolo]
@@ -1156,6 +1159,7 @@ anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 4
 classes=80
 num=9
 jitter=.1
+scale_x_y = 2.0
 objectness_smooth=0
 ignore_thresh = .7
 truth_thresh = 1
@@ -1169,6 +1173,7 @@ iou_loss=ciou
 nms_kind=diounms
 beta_nms=0.6
 new_coords=1
+max_delta=5
 
 [route]
 layers = -4
@@ -1275,7 +1280,7 @@ size=1
 stride=1
 pad=1
 filters=255
-activation=linear
+activation=logistic
 
 
 [yolo]
@@ -1284,6 +1289,7 @@ anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 4
 classes=80
 num=9
 jitter=.1
+scale_x_y = 2.0
 objectness_smooth=1
 ignore_thresh = .7
 truth_thresh = 1
@@ -1297,6 +1303,7 @@ iou_loss=ciou
 nms_kind=diounms
 beta_nms=0.6
 new_coords=1
+max_delta=5
 
 [route]
 layers = -4
@@ -1403,7 +1410,7 @@ size=1
 stride=1
 pad=1
 filters=255
-activation=linear
+activation=logistic
 
 
 [yolo]
@@ -1412,6 +1419,7 @@ anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 4
 classes=80
 num=9
 jitter=.1
+scale_x_y = 2.0
 objectness_smooth=1
 ignore_thresh = .7
 truth_thresh = 1
@@ -1425,3 +1433,4 @@ iou_loss=ciou
 nms_kind=diounms
 beta_nms=0.6
 new_coords=1
+max_delta=2
\ No newline at end of file
diff --git a/tests/darknet/yolo4-csp.cpp b/tests/darknet/yolo4-csp.cpp
new file mode 100644
index 0000000..3802a9a
--- /dev/null
+++ b/tests/darknet/yolo4-csp.cpp
@@ -0,0 +1,36 @@
+#include<iostream>
+#include<vector>
+#include "tkdnn.h"
+#include "test.h"
+#include "DarknetParser.h"
+
+int main() {
+    std::string bin_path  = "yolo4-csp";
+    std::vector<std::string> input_bins = {
+            bin_path + "/layers/input.bin"
+    };
+    std::vector<std::string> output_bins = {
+            bin_path + "/debug/layer144_out.bin",
+            bin_path + "/debug/layer159_out.bin",
+            bin_path + "/debug/layer174_out.bin"
+    };
+    std::string wgs_path  = bin_path + "/layers";
+    std::string cfg_path  = std::string(TKDNN_PATH) + "/tests/darknet/cfg/yolo4-csp.cfg";
+    std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/coco.names";
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/AfzHE4BfTeEm2gH/download");
+
+
+
+    // parse darknet network
+    tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
+    net->print();
+
+    //convert network to tensorRT
+    tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
+
+    int ret = testInference(input_bins, output_bins, net, netRT);
+    net->releaseLayers();
+    delete net;
+    delete netRT;
+    return ret;
+}
\ No newline at end of file
diff --git a/tests/darknet/yolo4x.cpp b/tests/darknet/yolo4x.cpp
index b9ad003..8df1aef 100644
--- a/tests/darknet/yolo4x.cpp
+++ b/tests/darknet/yolo4x.cpp
@@ -17,7 +17,7 @@ int main() {
     std::string wgs_path  = bin_path + "/layers";
     std::string cfg_path  = std::string(TKDNN_PATH) + "/tests/darknet/cfg/yolo4x.cfg";
     std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/coco.names";
-    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/BLPpiAigZJLorQD/download");
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/5MFjtNtgbDGdJEo/download");
 
 
 

From 7018d163ed2741bc1f095f9f129368beecdee567 Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Fri, 26 Mar 2021 19:37:44 +0530
Subject: [PATCH 38/53] python download file

---
 scripts/download_validation.py | 40 ++++++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)
 create mode 100644 scripts/download_validation.py

diff --git a/scripts/download_validation.py b/scripts/download_validation.py
new file mode 100644
index 0000000..e49f2ab
--- /dev/null
+++ b/scripts/download_validation.py
@@ -0,0 +1,40 @@
+import os
+from pathlib import Path 
+import urllib.request as dowReq
+import zipfile
+
+val = input("Enter BDD or COCO :")
+if(val == "COCO"):
+    url = "https://cloud.hipert.unimore.it/s/LNxBDk4wzqXPL8c/download"
+    lib = "..\demo\COCO_val2017"
+    lib_zip = "COCO_val2017.zip"
+elif(val == "BDD"):
+    url = "https://cloud.hipert.unimore.it/s/bikqk3FzCq2tg4D/download"
+    lib = "..\demo\BDD100k_val"
+    lib_zip = "BDD100k_val.zip"
+
+dowReq.urlretrieve(url,lib_zip)
+
+with zipfile.ZipFile(lib_zip,'r') as zip_ref:
+    zip_ref.extractall(lib)
+
+labelFolder = lib + "\labels"
+imageFolder = lib + "\images"
+
+file1 = open(".\\..\\demo\\all_labels.txt","a")
+path1 = os.path.realpath(labelFolder)
+for file in os.listdir(labelFolder):
+    valTemp = path1 + "\\" + file
+    valTemp = valTemp + " \n"
+    file1.write(valTemp)
+file1.close()
+
+file2 = open(".\\..\\demo\\all_images.txt","a")
+path2 = os.path.realpath(imageFolder)
+for file in os.listdir(imageFolder):
+    pathtemp = path2 + "\\" + file
+    pathtemp = pathtemp + " \n"
+    file2.write(pathtemp)
+file2.close()
+
+print("Completed")
\ No newline at end of file

From 5f3ab1472c8c02be43039abc273bcd6afc9f823d Mon Sep 17 00:00:00 2001
From: Harshvardhan Chandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Fri, 26 Mar 2021 15:09:28 +0100
Subject: [PATCH 39/53] Update download_validation.py

---
 scripts/download_validation.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/scripts/download_validation.py b/scripts/download_validation.py
index e49f2ab..e531b4f 100644
--- a/scripts/download_validation.py
+++ b/scripts/download_validation.py
@@ -1,5 +1,4 @@
 import os
-from pathlib import Path 
 import urllib.request as dowReq
 import zipfile
 
@@ -37,4 +36,4 @@ for file in os.listdir(imageFolder):
     file2.write(pathtemp)
 file2.close()
 
-print("Completed")
\ No newline at end of file
+print("Completed")

From 2e92944f1da276ec31eeb08bd99cf7e78679b23c Mon Sep 17 00:00:00 2001
From: hchandirasekar <f20180523@goa.bits-pilani.ac.in>
Date: Wed, 31 Mar 2021 03:44:51 -0700
Subject: [PATCH 40/53] Opencv cuda fix

---
 demo/demo/demo.cpp          | 2 +-
 include/tkDNN/DetectionNN.h | 4 +---
 2 files changed, 2 insertions(+), 4 deletions(-)

diff --git a/demo/demo/demo.cpp b/demo/demo/demo.cpp
index f97ead9..59bb1a2 100644
--- a/demo/demo/demo.cpp
+++ b/demo/demo/demo.cpp
@@ -25,7 +25,7 @@ int main(int argc, char *argv[]) {
     std::string net = "yolo4tiny_fp32.rt";
     if(argc > 1)
         net = argv[1]; 
-    std::string input = "../demo/yolo_test.mp4";
+    std::string input = "..\..\..\demo\yolo_test.mp4";
     if(argc > 2)
         input = argv[2]; 
     char ntype = 'y';
diff --git a/include/tkDNN/DetectionNN.h b/include/tkDNN/DetectionNN.h
index fb29be1..b1266e0 100644
--- a/include/tkDNN/DetectionNN.h
+++ b/include/tkDNN/DetectionNN.h
@@ -6,8 +6,6 @@
 #include <stdlib.h>    
 #ifdef __linux__
 #include <unistd.h>
-#elif _WIN32
-#include <Windows.h>
 #endif 
 
 #include <mutex>
@@ -19,7 +17,7 @@
 
 #include "tkdnn.h"
 
-// #define OPENCV_CUDACONTRIB //if OPENCV has been compiled with CUDA and contrib.
+#define OPENCV_CUDACONTRIB //if OPENCV has been compiled with CUDA and contrib.
 
 #ifdef OPENCV_CUDACONTRIB
 #include <opencv2/cudawarping.hpp>

From cc594f09efdbf0e7c5a33e4e4c40ea24ac2e7048 Mon Sep 17 00:00:00 2001
From: perseusdg <f20180523@goa.bits-pilani.ac.in>
Date: Thu, 8 Apr 2021 00:28:00 +0530
Subject: [PATCH 41/53] merge from gitlab

---
 include/tkDNN/NetworkRT.h | 10 ++++------
 include/tkDNN/utils.h     | 14 --------------
 src/NetworkRT.cpp         | 37 ++++---------------------------------
 src/utils.cpp             |  3 +--
 4 files changed, 9 insertions(+), 55 deletions(-)

diff --git a/include/tkDNN/NetworkRT.h b/include/tkDNN/NetworkRT.h
index b39360c..66892f5 100644
--- a/include/tkDNN/NetworkRT.h
+++ b/include/tkDNN/NetworkRT.h
@@ -55,16 +55,14 @@ class NetworkRT {
 
 public:
     nvinfer1::DataType dtRT;
-    //nvinfer1::IBuilder *builderRT;
-    std::unique_ptr<nvinfer1::IBuilder,InferDeleter> builderRT;
-    //nvinfer1::IRuntime *runtimeRT;
-    std::unique_ptr<nvinfer1::IRuntime,InferDeleter> runtimeRT;
+    nvinfer1::IBuilder *builderRT;
+    nvinfer1::IRuntime *runtimeRT;
     nvinfer1::INetworkDefinition *networkRT; 
 #if NV_TENSORRT_MAJOR >= 6  
     nvinfer1::IBuilderConfig *configRT;
 #endif
-    std::shared_ptr<nvinfer1::ICudaEngine> engineRT;
-   //nvinfer1::ICudaEngine *engineRT;
+    
+    nvinfer1::ICudaEngine *engineRT;
     nvinfer1::IExecutionContext *contextRT;
 
     const static int MAX_BUFFERS_RT = 10;
diff --git a/include/tkDNN/utils.h b/include/tkDNN/utils.h
index ce4a617..eeef3c2 100644
--- a/include/tkDNN/utils.h
+++ b/include/tkDNN/utils.h
@@ -14,9 +14,6 @@
 
 #ifdef __linux__
 #include <unistd.h>
-#elif _WIN32
-#define NOMINMAX
-#include <windows.h>
 #endif
 
 #include <ios>
@@ -110,17 +107,6 @@ double t_ns = time_ms.count();
       FatalError(_error.str());                                        \
     }                                                                  \
 }
-struct InferDeleter
-{
-    template <typename T>
-    void operator()(T* obj) const
-    {
-        if (obj)
-        {
-            obj->destroy();
-        }
-    }
-};
 
 typedef enum {
   ERROR_CUDNN = 2,
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index 1e0b062..b5005db 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -33,8 +33,7 @@ NetworkRT::NetworkRT(Network *net, const char *name) {
                    float(NV_TENSORRT_PATCH)/100;
     std::cout<<"New NetworkRT (TensorRT v"<<rt_ver<<")\n";
   
-   builderRT = std::unique_ptr<nvinfer1::IBuilder,InferDeleter>(createInferBuilder(loggerRT));
-    //builderRT = createInferBuilder(loggerRT);
+    builderRT = createInferBuilder(loggerRT);
     std::cout<<"Float16 support: "<<builderRT->platformHasFastFp16()<<"\n";
     std::cout<<"Int8 support: "<<builderRT->platformHasFastInt8()<<"\n";
 #if NV_TENSORRT_MAJOR >= 5
@@ -138,8 +137,7 @@ NetworkRT::NetworkRT(Network *net, const char *name) {
         printCudaMemUsage();
         std::cout<<"Building tensorRT cuda engine...\n";
 #if NV_TENSORRT_MAJOR >= 6                
-        //engineRT = builderRT->buildEngineWithConfig(*networkRT, *configRT);
-        engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildEngineWithConfig(*networkRT,*configRT),InferDeleter());
+        engineRT = builderRT->buildEngineWithConfig(*networkRT, *configRT);
 #else 
         //engineRT = builderRT->buildCudaEngine(*networkRT);
         engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildCudaEngine(*networkRT));
@@ -637,10 +635,8 @@ bool NetworkRT::deserialize(const char *filename) {
     }
 
     pluginFactory = new PluginFactory();
-    //runtimeRT = createInferRuntime(loggerRT);
-    runtimeRT = std::unique_ptr<nvinfer1::IRuntime,InferDeleter>(createInferRuntime(loggerRT));
-    //engineRT = runtimeRT->deserializeCudaEngine(gieModelStream, size, (IPluginFactory *) pluginFactory);
-    engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(runtimeRT->deserializeCudaEngine(gieModelStream,size,(IPluginFactory*)pluginFactory),InferDeleter());
+    runtimeRT = createInferRuntime(loggerRT);
+    engineRT = runtimeRT->deserializeCudaEngine(gieModelStream, size, (IPluginFactory *) pluginFactory);
     //if (gieModelStream) delete [] gieModelStream;
 
     return true;
@@ -673,7 +669,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
     }
     if(name.find("ActivationCReLU") == 0) {
         float activationReluTemp = readBUF<float>(buf);
-        //ActivationReLUCeiling *a = new ActivationReLUCeiling(readBUF<float>(buf));
         ActivationReLUCeiling* a = new ActivationReLUCeiling(activationReluTemp);
         a->size = readBUF<int>(buf);
         assert(buf == bufCheck + serialLength);
@@ -684,9 +679,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         int classesTemp = readBUF<int>(buf);
         int coordsTemp = readBUF<int>(buf);
         int numTemp = readBUF<int>(buf);
-        /*RegionRT *r = new RegionRT(readBUF<int>(buf),    //classes
-                                    readBUF<int>(buf),    //coords
-                                    readBUF<int>(buf));   //num8*/
         RegionRT* r = new RegionRT(classesTemp, coordsTemp, numTemp);
 
         r->c = readBUF<int>(buf);
@@ -698,7 +690,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
 
     if(name.find("Reorg") == 0) {
         int strideTemp = readBUF<int>(buf);
-        //ReorgRT *r = new ReorgRT(readBUF<int>(buf)); //stride
         ReorgRT *r = new ReorgRT(strideTemp);
         r->c = readBUF<int>(buf);
         r->h = readBUF<int>(buf);
@@ -723,15 +714,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
     } 
 
     if(name.find("Pooling") == 0) {
-       /* MaxPoolFixedSizeRT *r = new MaxPoolFixedSizeRT( readBUF<int>(buf), //c
-                                                        readBUF<int>(buf), //h
-                                                        readBUF<int>(buf), //w
-                                                        readBUF<int>(buf), //n
-                                                        readBUF<int>(buf), //strideH
-                                                        readBUF<int>(buf), //strideW
-                                                        readBUF<int>(buf), //winSize
-                                                        readBUF<int>(buf)); //padding*/
-
         int cTemp = readBUF<int>(buf);
         int hTemp = readBUF<int>(buf);
         int wTemp = readBUF<int>(buf);
@@ -747,9 +729,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
     }
 
     if(name.find("Resize") == 0) {
-        /*ResizeLayerRT *r = new ResizeLayerRT(readBUF<int>(buf), //o_c
-                                            readBUF<int>(buf), //o_h
-                                            readBUF<int>(buf)); //o_w*/
         int o_cTemp = readBUF<int>(buf);
         int o_hTemp = readBUF<int>(buf);
         int o_wTemp = readBUF<int>(buf);
@@ -822,7 +801,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
         return r;
     } 
     if(name.find("Upsample") == 0) {
-        //UpsampleRT *r = new UpsampleRT(readBUF<int>(buf)); //stride
         int strideTemp = readBUF<int>(buf);
         UpsampleRT* r = new UpsampleRT(strideTemp);
         r->c = readBUF<int>(buf);
@@ -833,7 +811,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
     } 
 
     if(name.find("Route") == 0) {
-        //RouteRT *r = new RouteRT(readBUF<int>(buf),readBUF<int>(buf));
         int groupsTemp = readBUF<int>(buf);
         int group_idTemp = readBUF<int>(buf);
         RouteRT* r = new RouteRT(groupsTemp, group_idTemp);
@@ -848,12 +825,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
     } 
 
     if(name.find("Deformable") == 0) {
-        /*DeformableConvRT *r = new DeformableConvRT(readBUF<int>(buf), readBUF<int>(buf), readBUF<int>(buf),
-                                                    readBUF<int>(buf), readBUF<int>(buf), readBUF<int>(buf),
-                                                    readBUF<int>(buf), readBUF<int>(buf),
-                                                    readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),
-                                                    readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),
-                                                    nullptr); */
         int chuck_dimTemp = readBUF<int>(buf);
         int khTemp = readBUF<int>(buf);
         int kwTemp = readBUF<int>(buf);
diff --git a/src/utils.cpp b/src/utils.cpp
index 52775df..fa6458f 100644
--- a/src/utils.cpp
+++ b/src/utils.cpp
@@ -203,8 +203,7 @@ void getMemUsage(double& vm_usage_kb, double& resident_set_kb){
 #ifdef __linux__
    long page_size_kb = sysconf(_SC_PAGE_SIZE) / 1024; // in case x86-64 is configured to use 2MB pages
 #elif _WIN32
-SYSTEM_INFO sysInfo;
- long page_size_kb = sysInfo.dwPageSize/1024;
+ long page_size_kb = 4096/1024;
 #endif
 
    vm_usage_kb     = vsize / 1024.0;  

From 37b2a5bd9856b8a6becb8ac696c95f70eaef20e9 Mon Sep 17 00:00:00 2001
From: perseusdg <f20180523@goa.bits-pilani.ac.in>
Date: Thu, 8 Apr 2021 00:57:58 +0530
Subject: [PATCH 42/53] minor modifications

---
 demo/demo/map.cpp               | 2 --
 include/tkDNN/ImuOdom.h         | 1 -
 include/tkDNN/Int8BatchStream.h | 2 --
 3 files changed, 5 deletions(-)

diff --git a/demo/demo/map.cpp b/demo/demo/map.cpp
index 8e363ee..58490e5 100644
--- a/demo/demo/map.cpp
+++ b/demo/demo/map.cpp
@@ -4,8 +4,6 @@
 #include <stdlib.h>     /* srand, rand */
 #ifdef __linux__
 #include <unistd.h>
-#elif _WIN32
-#include <Windows.h>
 #endif
 
 #include <mutex>
diff --git a/include/tkDNN/ImuOdom.h b/include/tkDNN/ImuOdom.h
index fa870f3..d5429a8 100644
--- a/include/tkDNN/ImuOdom.h
+++ b/include/tkDNN/ImuOdom.h
@@ -7,7 +7,6 @@
 #elif _WIN32
 #define _USE_MATH_DEFINES
 #include <math.h>
-#include <Windows.h>
 #endif 
 
 #include <mutex>
diff --git a/include/tkDNN/Int8BatchStream.h b/include/tkDNN/Int8BatchStream.h
index 7d2cef5..c39a11c 100644
--- a/include/tkDNN/Int8BatchStream.h
+++ b/include/tkDNN/Int8BatchStream.h
@@ -14,8 +14,6 @@
 #include <stdlib.h>
 #ifdef __linux__    
 #include <unistd.h>
-#elif _WIN32
-#include <Windows.h>
 #endif
 
 #include <mutex>

From 1de804f98dd67e66e6893e0c64749befdb9b6371 Mon Sep 17 00:00:00 2001
From: perseusdg <f20180523@goa.bits-pilani.ac.in>
Date: Fri, 9 Apr 2021 13:17:41 +0530
Subject: [PATCH 43/53] Code cleanup and readme fixes

---
 CMakeLists.txt                       |  6 ++---
 README.md                            | 39 +++++++++++++++++-----------
 demo/demo/demo.cpp                   |  7 ++++-
 include/tkDNN/DetectionNN.h          |  2 +-
 include/tkDNN/pluginsRT/UpsampleRT.h |  2 --
 include/tkDNN/pluginsRT/YoloRT.h     |  1 -
 scripts/download_validation.py       |  4 +--
 src/NetworkRT.cpp                    |  2 +-
 8 files changed, 37 insertions(+), 26 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 1b7ed63..c27e519 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -3,11 +3,11 @@ cmake_minimum_required(VERSION 3.5)
 project (tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 if(UNIX)
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable -g ")
 endif()
 if(WIN32)
-set(CMAKE_CXX_STANDARD 14)
-set(CMAKE_CXX_FLAGS "/O1 /FS /EHsc")
+set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CXX_FLAGS "/O2 /FS /EHsc")
 set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
 endif(WIN32)
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include/tkDNN)
diff --git a/README.md b/README.md
index f685b58..cbc65b0 100644
--- a/README.md
+++ b/README.md
@@ -80,12 +80,13 @@ Results for COCO val 2017 (5k images), on RTX 2080Ti, with conf threshold=0.001
   - [mAP demo](#map-demo)
   - [Existing tests and supported networks](#existing-tests-and-supported-networks)
   - [References](#references)
-  - [tkDNN on Windows 10 (experimental)](#tkdnn-on-windows)
+  - [tkDNN on Windows 10 (experimental)](#tkdnn-on-windows-10-experimental)
     - [Dependencies-Windows](#dependencies-windows)
-    - [Compiling tkDNN on Windows](#tkdnn-windows-compile)
+    - [Compiling tkDNN on Windows](#compiling-tkdnn-on-windows)
     - [Run the demo on Windows](#run-the-demo-on-windows)
-    - [FP16 interference windows](#fp16-windows)
-    - [INT8 interference windows](#int8-windows)
+      - [FP16 inference windows](#fp16-inference-windows)
+      - [INT8 inference windows](#int8-inference-windows)
+    - [Known issues with tkDNN on Windows](#known-issues-with-tkdnn-on-windows)
   
 
 
@@ -362,26 +363,31 @@ This demo also creates a json file named ```net_name_COCO_res.json``` containing
 | yolo4tiny             | Yolov4 tiny <sup>9</sup>                           | [COCO 2017](http://cocodataset.org/)                          | 80        | 416x416       | [weights](https://cloud.hipert.unimore.it/s/iRnc4pSqmx78gJs/download)     |
 | yolo4x             | Yolov4x-mish  <sup>9</sup>                          | [COCO 2017](http://cocodataset.org/)                          | 80        | 672x672       | [weights](https://cloud.hipert.unimore.it/s/BLPpiAigZJLorQD/download)     |
 
-##tkDNN on Windows 10 (experimental)
+### tkDNN on Windows 10 (experimental)
 
 ### Dependencies-Windows 
 This branch should work on every NVIDIA GPU supported in windows with the following dependencies:
 
 * WINDOWS 10 1803 or HIGHER 
-* CUDA 10.0 (Recommended CUDA 11.0 +)
-* CUDNN 7.6 (Recommended CUDNN 8.0.0 +)
-* TENSORRT 6.0.1 (Recommended TENSORRT 7.1 +)
-* OPENCV 3.4 (Recommended OPENCV 4.2.0 +)
-* MSVC 16.7 (Recommended MSVC 16.8/16.9)
-* YAML-CPP 0.5.2 
+* CUDA 10.0 (Recommended CUDA 11.2 )
+* CUDNN 7.6 (Recommended CUDNN 8.1.1 )
+* TENSORRT 6.0.1 (Recommended TENSORRT 7.2.3.4 )
+* OPENCV 3.4 (Recommended OPENCV 4.2.0 )
+* MSVC 16.7 
+* YAML-CPP 
 * EIGEN3
 * 7ZIP (ADD TO PATH)
 * NINJA 1.10
 
+
 All the above mentioned dependencies except 7ZIP can be installed using Microsoft's [VCPKG](https://github.com/microsoft/vcpkg.git) .
 After bootstrapping VCPKG the dependencies can be built and installed using the following command :
 
-```vcpkg.exe install opencv4[tbb,jpeg,tiff,opengl,openmp,png,ffmpeg]:x64-windows yaml-cpp:x64-windows eigen3:x64-windows --x-install-root=C:\opt --x-buildtrees-root=C:\temp_vcpkg_build```
+```
+opencv4(normal) - vcpkg.exe install opencv4[tbb,jpeg,tiff,opengl,openmp,png,ffmpeg,eigen]:x64-windows yaml-cpp:x64-windows eigen3:x64-windows --x-install-root=C:\opt --x-buildtrees-root=C:\temp_vcpkg_build
+
+opencv4(cuda) - vcpkg.exe install opencv4[cuda,nonfree,contrib,eigen,tbb,jpeg,tiff,opengl,openmp,png,ffmpeg]:x64-windows yaml-cpp:x64-windows eigen3:x64-windows --x-install-root=C:\opt --x-buildtrees-root=C:\temp_vcpkg_build
+```
 
 After VCPKG finishes building and installing all the packages delete C:\temp_vcpkg_build and add C:\opt\x64-windows\bin and C:\opt\x64-windows\debug\bin to path 
 
@@ -411,7 +417,7 @@ Once the rt file has been successfully create,run the demo using the following c
 ```
  For general info on more demo paramters,check Run the demo section on top 
 
-### FP16 interference windows 
+### FP16 inference windows 
 
 This is an untested feature on windows.To run the object detection demo with FP16 interference follow the below steps(example with yolo4tiny):
 ```
@@ -421,7 +427,7 @@ del /f yolo4tiny_fp16.rt
 .\demo.exe yolo4tiny_fp16.rt ..\demo\yolo_test.mp4
 ```
 
-### INT8 interference windows 
+### INT8 inference windows 
 To run object detection demo with INT8 (example with yolo4tiny):
 ```
 set TKDNN_MODE=INT8
@@ -433,10 +439,13 @@ del /f  yolo4tiny_int8.rt        # be sure to delete(or move) old tensorRT files
 
 ```
 
+### Known issues with tkDNN on Windows
 
+Mobilenet and Centernet demos work properly only when built with msvc 16.7 in Release Mode,when built in debug mode for the mentioned networks one might encounter opencv assert errors
 
+All Darknet models work properly with demo using MSVC version(16.7-16.9)
 
-
+It is recommended to use Nvidia Driver(465+),Cuda unknown errors have been observed when using older drivers on pascal(SM 61) devices.
 
 
 
diff --git a/demo/demo/demo.cpp b/demo/demo/demo.cpp
index 59bb1a2..317a574 100644
--- a/demo/demo/demo.cpp
+++ b/demo/demo/demo.cpp
@@ -25,7 +25,12 @@ int main(int argc, char *argv[]) {
     std::string net = "yolo4tiny_fp32.rt";
     if(argc > 1)
         net = argv[1]; 
-    std::string input = "..\..\..\demo\yolo_test.mp4";
+    #ifdef __linux__ 
+        std::string input = "../demo/yolo_test.mp4";
+    #elif _WIN32
+        std::string input = "..\\..\\..\\demo\\yolo_test.mp4";
+    #endif
+
     if(argc > 2)
         input = argv[2]; 
     char ntype = 'y';
diff --git a/include/tkDNN/DetectionNN.h b/include/tkDNN/DetectionNN.h
index b1266e0..a8c81f7 100644
--- a/include/tkDNN/DetectionNN.h
+++ b/include/tkDNN/DetectionNN.h
@@ -17,7 +17,7 @@
 
 #include "tkdnn.h"
 
-#define OPENCV_CUDACONTRIB //if OPENCV has been compiled with CUDA and contrib.
+//#define OPENCV_CUDACONTRIB //if OPENCV has been compiled with CUDA and contrib.
 
 #ifdef OPENCV_CUDACONTRIB
 #include <opencv2/cudawarping.hpp>
diff --git a/include/tkDNN/pluginsRT/UpsampleRT.h b/include/tkDNN/pluginsRT/UpsampleRT.h
index 5350b7e..a11d7b4 100644
--- a/include/tkDNN/pluginsRT/UpsampleRT.h
+++ b/include/tkDNN/pluginsRT/UpsampleRT.h
@@ -59,8 +59,6 @@ public:
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
-		std::cout << "Upsample Serialization SIze" << getSerializationSize() << std::endl;
-
 		assert(buf == a + getSerializationSize());
 	}
 
diff --git a/include/tkDNN/pluginsRT/YoloRT.h b/include/tkDNN/pluginsRT/YoloRT.h
index 0dd26e1..2911869 100644
--- a/include/tkDNN/pluginsRT/YoloRT.h
+++ b/include/tkDNN/pluginsRT/YoloRT.h
@@ -120,7 +120,6 @@ public:
 				tk::dnn::writeBUF(buf, tmp[j]);
 			}
 		}
-		std::cout << getSerializationSize() << std::endl;
 		assert(buf == a + getSerializationSize());
 	}
 
diff --git a/scripts/download_validation.py b/scripts/download_validation.py
index e531b4f..0e3b1d4 100644
--- a/scripts/download_validation.py
+++ b/scripts/download_validation.py
@@ -24,7 +24,7 @@ file1 = open(".\\..\\demo\\all_labels.txt","a")
 path1 = os.path.realpath(labelFolder)
 for file in os.listdir(labelFolder):
     valTemp = path1 + "\\" + file
-    valTemp = valTemp + " \n"
+    valTemp = valTemp + '\n'
     file1.write(valTemp)
 file1.close()
 
@@ -32,7 +32,7 @@ file2 = open(".\\..\\demo\\all_images.txt","a")
 path2 = os.path.realpath(imageFolder)
 for file in os.listdir(imageFolder):
     pathtemp = path2 + "\\" + file
-    pathtemp = pathtemp + " \n"
+    pathtemp = pathtemp + '\n'
     file2.write(pathtemp)
 file2.close()
 
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index b5005db..5dc8ee0 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -648,7 +648,7 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
     const char * buf = reinterpret_cast<const char*>(serialData),*bufCheck = buf;
 
     std::string name(layerName);
-    std::cout<<name<<std::endl;
+    //std::cout<<name<<std::endl;
 
     if(name.find("ActivationLeaky") == 0) {
         ActivationLeakyRT *a = new ActivationLeakyRT();

From f90ee8ab7df5a366164d346bea403b1d5be9c3e6 Mon Sep 17 00:00:00 2001
From: Harshvardhan Chandirasekar <43143075+perseusdg@users.noreply.github.com>
Date: Fri, 9 Apr 2021 13:20:44 +0530
Subject: [PATCH 44/53] Update CMakeLists.txt

---
 CMakeLists.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index c27e519..d3a89f5 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -3,7 +3,7 @@ cmake_minimum_required(VERSION 3.5)
 project (tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 if(UNIX)
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable -g ")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable ")
 endif()
 if(WIN32)
 set(CMAKE_CXX_STANDARD 11)

From 39323ca8d3ac02a8a1e74ede6c851e57546f18ab Mon Sep 17 00:00:00 2001
From: Harshvardhan Chandirasekar <43143075+perseusdg@users.noreply.github.com>
Date: Wed, 14 Apr 2021 17:22:26 +0530
Subject: [PATCH 45/53] Update README.md

---
 README.md | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/README.md b/README.md
index cbc65b0..7fb389b 100644
--- a/README.md
+++ b/README.md
@@ -388,6 +388,7 @@ opencv4(normal) - vcpkg.exe install opencv4[tbb,jpeg,tiff,opengl,openmp,png,ffmp
 
 opencv4(cuda) - vcpkg.exe install opencv4[cuda,nonfree,contrib,eigen,tbb,jpeg,tiff,opengl,openmp,png,ffmpeg]:x64-windows yaml-cpp:x64-windows eigen3:x64-windows --x-install-root=C:\opt --x-buildtrees-root=C:\temp_vcpkg_build
 ```
+To build opencv4 with cuda and cudnn version corresponding to your cuda version,vcpkg's cudnn portfile needs to be modified by adding ```$ENV{CUDA_PATH}```  at lines 16 and 17 in the portfile.cmake 
 
 After VCPKG finishes building and installing all the packages delete C:\temp_vcpkg_build and add C:\opt\x64-windows\bin and C:\opt\x64-windows\debug\bin to path 
 
@@ -395,7 +396,7 @@ After VCPKG finishes building and installing all the packages delete C:\temp_vcp
 
 tkDNN is built with cmake(3.15+) on windows along with ninja.Msbuild and NMake Makefiles are drastically slower when compiling the library compared to windows
 ```
-git clone https://git.hipert.unimore.it/research-cv-chandirasekar/tkdnn-windows.git
+git clone https://github.com/ceccocats/tkDNN.git
 cd tkdnn-windows
 mkdir build
 cd build
@@ -416,6 +417,7 @@ Once the rt file has been successfully create,run the demo using the following c
 .\demo.exe yolo4tiny_fp32.rt ..\demo\yolo_test.mp4 y 
 ```
  For general info on more demo paramters,check Run the demo section on top 
+ To run the test_all_tests.sh on windows,use git bash or msys2 
 
 ### FP16 inference windows 
 

From 6b8ae1e27c54abd61f68dd8d98f903386fb9b391 Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Sun, 16 May 2021 22:16:21 +0200
Subject: [PATCH 46/53] remove wrong cuda arch that cause big performance gap

fix #226
---
 CMakeLists.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index d3a89f5..0383add 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -25,7 +25,7 @@ add_definitions(-DTKDNN_PATH="${CMAKE_CURRENT_SOURCE_DIR}")
 find_package(CUDA 9.0 REQUIRED)
 SET(CUDA_SEPARABLE_COMPILATION ON)
 #set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -arch=sm_30 --compiler-options '-fPIC'")
-set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -arch=sm_61 )
+set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32)
 
 find_package(CUDNN REQUIRED)
 include_directories(${CUDNN_INCLUDE_DIR})

From ba8199a03088b7c8e36066a1636a1237ab316cec Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Sun, 16 May 2021 22:33:45 +0200
Subject: [PATCH 47/53] fix compile error on old tensorrt and remove useless
 prints

---
 include/tkDNN/pluginsRT/YoloRT.h | 24 ++++++++++++------------
 src/NetworkRT.cpp                |  4 ++--
 2 files changed, 14 insertions(+), 14 deletions(-)

diff --git a/include/tkDNN/pluginsRT/YoloRT.h b/include/tkDNN/pluginsRT/YoloRT.h
index 2911869..5ffe39c 100644
--- a/include/tkDNN/pluginsRT/YoloRT.h
+++ b/include/tkDNN/pluginsRT/YoloRT.h
@@ -93,23 +93,23 @@ public:
 
 	virtual void serialize(void* buffer) override {
 		char *buf = reinterpret_cast<char*>(buffer),*a=buf;
-		tk::dnn::writeBUF(buf, classes); std::cout << "Classes :" << classes << std::endl;
-		tk::dnn::writeBUF(buf, num); std::cout << "Num : " << num << std::endl;
-		tk::dnn::writeBUF(buf, n_masks); std::cout << "N_Masks" << n_masks << std::endl;
-		tk::dnn::writeBUF(buf, scaleXY); std::cout << "ScaleXY :" << scaleXY << std::endl;
-		tk::dnn::writeBUF(buf, nms_thresh); std::cout << "nms_thresh :" << nms_thresh << std::endl;
-		tk::dnn::writeBUF(buf, nms_kind); std::cout << "nms_kind : " << nms_kind << std::endl;
-		tk::dnn::writeBUF(buf, new_coords); std::cout << "new_coords : " << new_coords << std::endl;
-		tk::dnn::writeBUF(buf, c); std::cout << "C : " << c << std::endl;
-		tk::dnn::writeBUF(buf, h); std::cout << "H : " << h << std::endl;
-		tk::dnn::writeBUF(buf, w); std::cout << "C : " << c << std::endl;
+		tk::dnn::writeBUF(buf, classes); 	//std::cout << "Classes :" << classes << std::endl;
+		tk::dnn::writeBUF(buf, num); 		//std::cout << "Num : " << num << std::endl;
+		tk::dnn::writeBUF(buf, n_masks); 	//std::cout << "N_Masks" << n_masks << std::endl;
+		tk::dnn::writeBUF(buf, scaleXY); 	//std::cout << "ScaleXY :" << scaleXY << std::endl;
+		tk::dnn::writeBUF(buf, nms_thresh); //std::cout << "nms_thresh :" << nms_thresh << std::endl;
+		tk::dnn::writeBUF(buf, nms_kind); 	//std::cout << "nms_kind : " << nms_kind << std::endl;
+		tk::dnn::writeBUF(buf, new_coords); //std::cout << "new_coords : " << new_coords << std::endl;
+		tk::dnn::writeBUF(buf, c); 			//std::cout << "C : " << c << std::endl;
+		tk::dnn::writeBUF(buf, h); 			//std::cout << "H : " << h << std::endl;
+		tk::dnn::writeBUF(buf, w); 			//std::cout << "C : " << c << std::endl;
 		for (int i = 0; i < n_masks; i++)
 		{
-			tk::dnn::writeBUF(buf, mask[i]); std::cout << "mask[i] : " << mask[i] << std::endl;
+			tk::dnn::writeBUF(buf, mask[i]); //std::cout << "mask[i] : " << mask[i] << std::endl;
 		}
 		for (int i = 0; i < n_masks * 2 * num; i++)
 		{
-			tk::dnn::writeBUF(buf, bias[i]); std::cout << "bias[i] : " << bias[i] << std::endl;
+			tk::dnn::writeBUF(buf, bias[i]); //std::cout << "bias[i] : " << bias[i] << std::endl;
 		}
 
 		// save classes names
diff --git a/src/NetworkRT.cpp b/src/NetworkRT.cpp
index 320c3ea..a95423d 100644
--- a/src/NetworkRT.cpp
+++ b/src/NetworkRT.cpp
@@ -139,8 +139,8 @@ NetworkRT::NetworkRT(Network *net, const char *name) {
 #if NV_TENSORRT_MAJOR >= 6                
         engineRT = builderRT->buildEngineWithConfig(*networkRT, *configRT);
 #else 
-        //engineRT = builderRT->buildCudaEngine(*networkRT);
-        engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildCudaEngine(*networkRT));
+        engineRT = builderRT->buildCudaEngine(*networkRT);
+        //engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildCudaEngine(*networkRT));
 #endif
         if(engineRT == nullptr)
             FatalError("cloud not build cuda engine")

From b0fdeb41275cf320c8ab2fea379da07a2bd35f08 Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Tue, 15 Jun 2021 19:38:13 +0200
Subject: [PATCH 48/53] install targets

---
 CMakeLists.txt | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 0383add..48590d9 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -85,6 +85,7 @@ foreach(test_SRC ${darknet_SRC})
     set(test_NAME test_${test_NAME})
     add_executable(${test_NAME} ${test_SRC})
     target_link_libraries(${test_NAME} tkDNN)
+    install(TARGETS ${test_NAME} DESTINATION bin)
 endforeach()
 
 # MOBILENET
@@ -131,6 +132,7 @@ target_link_libraries(demo tkDNN)
 message("install dir:" ${CMAKE_INSTALL_PREFIX})
 install(DIRECTORY include/ DESTINATION include/)
 install(TARGETS tkDNN kernels DESTINATION lib)
+install(TARGETS test_rtinference demo map_demo DESTINATION bin)
 install(DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/cmake/" # source directory
         DESTINATION "share/tkDNN/cmake/" # target directory
 )

From 6d7c456f724c7456afa3c594ddb72d90e6be4926 Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Tue, 15 Jun 2021 19:39:34 +0200
Subject: [PATCH 49/53] cmake version fix

---
 CMakeLists.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 48590d9..dd900bd 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.5)
+cmake_minimum_required(VERSION 3.15)
 
 project (tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)

From 8df7d5fd1b55b99e111404f8d06eebb1017e5237 Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Tue, 15 Jun 2021 20:31:58 +0200
Subject: [PATCH 50/53] Update CMakeLists.txt

---
 CMakeLists.txt | 1 +
 1 file changed, 1 insertion(+)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index dd900bd..f09375a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -42,6 +42,7 @@ target_link_libraries(kernels ${CUDA_CUBLAS_LIBRARIES})
 # External Libraries
 #-------------------------------------------------------------------------------
 find_package(Eigen3 REQUIRED)
+include_directories("Eigen DIR: " ${EIGEN3_INCLUDE_DIR})
 include_directories(${EIGEN3_INCLUDE_DIR})
 
 find_package(OpenCV REQUIRED)

From d847a4b8520e7887f40ccfb3cbcd349c19eb08dc Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Tue, 15 Jun 2021 20:33:41 +0200
Subject: [PATCH 51/53] Update CMakeLists.txt

---
 CMakeLists.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index f09375a..4d67e1b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -42,7 +42,7 @@ target_link_libraries(kernels ${CUDA_CUBLAS_LIBRARIES})
 # External Libraries
 #-------------------------------------------------------------------------------
 find_package(Eigen3 REQUIRED)
-include_directories("Eigen DIR: " ${EIGEN3_INCLUDE_DIR})
+message("Eigen DIR: " ${EIGEN3_INCLUDE_DIR})
 include_directories(${EIGEN3_INCLUDE_DIR})
 
 find_package(OpenCV REQUIRED)

From 9b78f143cb3da2392d22b32383b0d8ea4de854ce Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Wed, 16 Jun 2021 14:53:46 +0200
Subject: [PATCH 52/53] Update CMakeLists.txt

---
 CMakeLists.txt | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 4d67e1b..6076459 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -17,7 +17,13 @@ if(DEBUG)
     add_definitions(-DDEBUG)
 endif()
 
-add_definitions(-DTKDNN_PATH="${CMAKE_CURRENT_SOURCE_DIR}")
+if(TKDNN_PATH)
+    message("SET TKDNN_PATH:"${TKDNN_PATH})
+    add_definitions(-DTKDNN_PATH="${TKDNN_PATH}")
+else()
+    add_definitions(-DTKDNN_PATH="${CMAKE_CURRENT_SOURCE_DIR}")
+endif()
+
 
 #-------------------------------------------------------------------------------
 # CUDA
@@ -137,4 +143,6 @@ install(TARGETS test_rtinference demo map_demo DESTINATION bin)
 install(DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/cmake/" # source directory
         DESTINATION "share/tkDNN/cmake/" # target directory
 )
-
+install(DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/tests/" # source directory
+        DESTINATION "share/tkDNN/tests" # target directory
+)

From c306b368608893e92925bf143e7cf14f19525aeb Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Thu, 17 Jun 2021 15:31:25 +0200
Subject: [PATCH 53/53] Update CMakeLists.txt

---
 CMakeLists.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 6076459..f03c98c 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -139,7 +139,7 @@ target_link_libraries(demo tkDNN)
 message("install dir:" ${CMAKE_INSTALL_PREFIX})
 install(DIRECTORY include/ DESTINATION include/)
 install(TARGETS tkDNN kernels DESTINATION lib)
-install(TARGETS test_rtinference demo map_demo DESTINATION bin)
+install(TARGETS test_simple test_mnist test_mnistRT test_rtinference demo map_demo DESTINATION bin)
 install(DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/cmake/" # source directory
         DESTINATION "share/tkDNN/cmake/" # target directory
 )