darknet parser tested

include/tkDNN/DarknetParser.h

@@ -22,6 +22,7 @@ namespace tk { namespace dnn {
         int classes = 20;
         int num = 1;
         int pad = 0;
+        int coords = 4;
         float scale_xy = 1;
         std::vector<int> layers;
         std::string activation = "linear";
@@ -102,9 +103,11 @@ namespace tk { namespace dnn {
             fields.classes = std::stoi(value);
         else if(name.find("num") !=  std::string::npos)
             fields.num = std::stoi(value);
+        else if(name.find("coords") !=  std::string::npos)
+            fields.coords = std::stoi(value);
         else if(name.find("groups") !=  std::string::npos)
             fields.groups = std::stoi(value);
-        else if(name.find("scale_xy") !=  std::string::npos)
+        else if(name.find("scale_x_y") !=  std::string::npos)
             fields.scale_xy = std::stof(value);
         else if(name.find("from") !=  std::string::npos)
             fields.layers.push_back(std::stof(value));
@@ -135,23 +138,25 @@ namespace tk { namespace dnn {
         if(f.pad == 1) {
             f.padding_x = f.padding_y = f.size_x /2;
         }
-
         std::cout<<"Add layer: "<<f.type<<"\n";
         if(f.type == "convolutional") {
             std::string wgs = wgs_path + "/c" + std::to_string(netLayers.size()) + ".bin";
             printf("%d (%d,%d) (%d,%d) (%d,%d) %s %d %d\n", f.filters, f.size_x, f.size_y, f.stride_x, f.stride_y, f.padding_x, f.padding_y, wgs.c_str(), f.batch_normalize, f.groups);
             tk::dnn::Conv2d *l= new tk::dnn::Conv2d(net, f.filters, f.size_x, f.size_y, f.stride_x, 
                                 f.stride_y, f.padding_x, f.padding_y, wgs, f.batch_normalize, false, f.groups);
-            if(f.activation != "linear") {
+            netLayers.push_back(l);
-                tkdnnActivationMode_t act;
+        } else if(f.type == "maxpool") {
-                if(f.activation == "relu") act = tkdnnActivationMode_t(CUDNN_ACTIVATION_RELU);
+            if(f.stride_x == 1 && f.stride_y == 1)
-                else if(f.activation == "leaky") act = tk::dnn::ACTIVATION_LEAKY;
+                netLayers.push_back(new tk::dnn::Pooling(net, f.size_x, f.size_y, f.stride_x, f.stride_y, 
-                else if(f.activation == "mish") act = tk::dnn::ACTIVATION_MISH;
+                    f.padding_x, f.padding_y, tk::dnn::POOLING_MAX_FIXEDSIZE));
-                else { FatalError("activation not supported: " + f.activation); }
+            else
-                netLayers.push_back(new tk::dnn::Activation(net, act));
+                netLayers.push_back(new tk::dnn::Pooling(net, f.size_x, f.size_y, f.stride_x, f.stride_y, 
-            } else {
+                    f.padding_x, f.padding_y, tk::dnn::POOLING_MAX));
-                netLayers.push_back(l);
+
-            }
+        } else if(f.type == "avgpool") {
+            netLayers.push_back(new tk::dnn::Pooling(net, f.size_x, f.size_y, f.stride_x, f.stride_y, 
+                f.padding_x, f.padding_y, tk::dnn::POOLING_AVERAGE));
+
         } else if(f.type == "shortcut") {
             if(f.layers.size() != 1) FatalError("no layers to shortcut\n");
             int layerIdx = f.layers[0];
@@ -177,6 +182,12 @@ namespace tk { namespace dnn {
             }
             netLayers.push_back(new tk::dnn::Route(net, layers.data(), layers.size()));
 
+        } else if(f.type == "reorg") {
+            netLayers.push_back(new tk::dnn::Reorg(net, f.stride_x));
+
+        } else if(f.type == "region") {
+            netLayers.push_back(new tk::dnn::Region(net, f.classes, f.coords, f.num));
+
         } 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);
@@ -189,6 +200,16 @@ namespace tk { namespace dnn {
         } else{
             FatalError("layer not supported: " + f.type);
         }
+
+        // add activation
+        if(netLayers.size() > 0 && f.activation != "linear") {
+            tkdnnActivationMode_t act;
+            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 { FatalError("activation not supported: " + f.activation); }
+            netLayers[netLayers.size()-1] = new tk::dnn::Activation(net, act);
+        };
     }
 
     std::vector<std::string> darknetReadNames(const std::string& names_file){
@@ -244,7 +265,7 @@ namespace tk { namespace dnn {
 
                 // new type
                 //std::cout<<"type: "<<type<<"\n";
-                fields = darknetFields_t();
+                fields = darknetFields_t(); // reset to default
                 fields.type = type;
                 continue;
             }

include/tkDNN/Network.h

@@ -40,6 +40,7 @@ class Network {
 public:
     Network(dataDim_t input_dim);
     virtual ~Network();
+    void releaseLayers();
 
     /**
         Do inferece for every added layer

include/tkDNN/test.h

@@ -8,6 +8,11 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
         if(net->layers[i]->final)
             outputs.push_back(net->layers[i]);
     }
+    // no final layers, set last as output
+    if(outputs.size() == 0) {
+        outputs.push_back(net->layers[net->num_layers-1]);
+    }
+
 
     // check input
     if(input_bins.size() != 1) {

scripts/test_all_tests.sh

@@ -78,9 +78,10 @@ do
     test_net yolo3_512
     test_net yolo3tiny
     test_net csresnext50-panet-spp
+    #test_net csresnext50-panet-spp_berkeley
     test_net mobilenetv2ssd
     test_net yolo3tiny_512
-    test_net yolo2tiny
+    #test_net yolo2tiny
     test_net mobilenetv2ssd512
     test_net mnist
     test_net yolo2

src/Network.cpp

@@ -59,12 +59,16 @@ Network::Network(dataDim_t input_dim) {
 }
 
 Network::~Network() {
-    for(int i=0; i<num_layers; i++)
-        delete layers[i];
     checkCUDNN( cudnnDestroy(cudnnHandle) );
     checkERROR( cublasDestroy(cublasHandle) );
 }
 
+void Network::releaseLayers() {
+    for(int i=0; i<num_layers; i++)
+        delete layers[i];
+    num_layers = 0;
+}
+
 dnnType* Network::infer(dataDim_t &dim, dnnType* data) {
 
     //do infer for every layer

src/Region.cpp

@@ -12,8 +12,7 @@
 namespace tk { namespace dnn {
 
 Region::Region(Network *net, int classes, int coords, int num) : 
-    Layer(net) {
+    Layer(net) {    
-
     this->classes = classes;
     this->coords = coords;
     this->num = num;

tests/darknet/cfg/yolo2tiny.cfg

@@ -1,5 +1,5 @@
 [net]
- Training
+# Training
  batch=64
  subdivisions=8
 # Testing
@@ -90,9 +90,9 @@ stride=1
 pad=1
 activation=leaky
 
-#[maxpool]
+[maxpool]
-#size=2
+size=2
-#stride=1
+stride=1
 
 [convolutional]
 batch_normalize=1

tests/darknet/csresnext50-panet-spp.cpp

@@ -27,6 +27,7 @@ int main() {
     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;

tests/darknet/csresnext50-panet-spp_berkeley.cpp

@@ -5,7 +5,7 @@
 #include "DarknetParser.h"
 
 int main() {
-    std::string bin_path = "bdd-csresnext50-panet-spp";
+    std::string bin_path = "csresnext50-panet-spp_berkeley";
     std::vector<std::string> input_bins = { 
         bin_path + "/layers/input.bin"
     };
@@ -17,6 +17,7 @@ int main() {
     std::string wgs_path  = bin_path + "/layers";
     std::string cfg_path  = "../tests/darknet/cfg/csresnext50-panet-spp_berkeley.cfg";
     std::string name_path = "../tests/darknet/names/berkeley.names";
+    // FIXME: wrong weights
     // downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s//download");
 
     // parse darknet network
@@ -27,6 +28,7 @@ int main() {
     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;

tests/darknet/yolo2.cpp

@@ -10,7 +10,7 @@ int main() {
         bin_path + "/layers/input.bin"
     };
     std::vector<std::string> output_bins = {
-        bin_path + "layers/output.bin"
+        bin_path + "/layers/output.bin"
     };
     std::string wgs_path  = bin_path + "/layers";
     std::string cfg_path  = "../tests/darknet/cfg/yolo2.cfg";
@@ -25,6 +25,7 @@ int main() {
     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;

tests/darknet/yolo2_voc.cpp

@@ -25,6 +25,7 @@ int main() {
     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;

tests/darknet/yolo2tiny.cpp

@@ -10,12 +10,13 @@ int main() {
         bin_path + "/layers/input.bin"
     };
     std::vector<std::string> output_bins = {
-        bin_path + "layers/output.bin"
+        bin_path + "/layers/output.bin"
     };
     std::string wgs_path  = bin_path + "/layers";
     std::string cfg_path  = "../tests/darknet/cfg/yolo2tiny.cfg";
     std::string name_path = "../tests/darknet/names/coco.names";
-    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/nf4PJ3k8bxBETwL/download");
+    // FIXME: wrong weights
+    //downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s//download");
 
     // parse darknet network
     tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
@@ -25,6 +26,7 @@ int main() {
     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;

tests/darknet/yolo3.cpp

@@ -26,7 +26,8 @@ int main() {
     //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);
+    int ret = testInference(input_bins, output_bins, net, netRT);    
+    net->releaseLayers();
     delete net;
     delete netRT;
     return ret;

tests/darknet/yolo3_512.cpp

@@ -27,6 +27,7 @@ int main() {
     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;

tests/darknet/yolo3_berkeley.cpp

@@ -16,7 +16,7 @@ int main() {
     };
     std::string wgs_path  = bin_path + "/layers";
     std::string cfg_path  = "../tests/darknet/cfg/yolo3_berkeley.cfg";
-    std::string name_path = "../tests/darknet/names/barkeley.names";
+    std::string name_path = "../tests/darknet/names/berkeley.names";
     downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/o5cHa4AjTKS64oD/download");
 
     // parse darknet network
@@ -27,6 +27,7 @@ int main() {
     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;

tests/darknet/yolo3_coco4.cpp

@@ -27,6 +27,7 @@ int main() {
     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;

tests/darknet/yolo3_flir.cpp

@@ -27,6 +27,7 @@ int main() {
     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;

tests/darknet/yolo3tiny.cpp

@@ -10,7 +10,8 @@ int main() {
         bin_path + "/layers/input.bin"
     };
     std::vector<std::string> output_bins = {
-        bin_path + "debug/layer23_out.bin",
+        bin_path + "/debug/layer16_out.bin",
+        bin_path + "/debug/layer23_out.bin",
     };
     std::string wgs_path  = bin_path + "/layers";
     std::string cfg_path  = "../tests/darknet/cfg/yolo3tiny.cfg";
@@ -25,6 +26,7 @@ int main() {
     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;

tests/darknet/yolo3tiny_512.cpp

@@ -10,7 +10,8 @@ int main() {
         bin_path + "/layers/input.bin"
     };
     std::vector<std::string> output_bins = {
-        bin_path + "debug/layer23_out.bin",
+        bin_path + "/debug/layer16_out.bin",
+        bin_path + "/debug/layer23_out.bin",
     };
     std::string wgs_path  = bin_path + "/layers";
     std::string cfg_path  = "../tests/darknet/cfg/yolo3tiny_512.cfg";
@@ -25,6 +26,7 @@ int main() {
     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;

tests/darknet/yolo4.cpp

@@ -27,6 +27,7 @@ int main() {
     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;

tests/darknet/yolo4_berkeley.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_berkeley";
+    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  = "../tests/darknet/cfg/yolo4_berkeley.cfg";
+    std::string name_path = "../tests/darknet/names/berkeley.names";
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s//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;
+}