Add different yolov4 size tests

Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>

include/tkDNN/utils.h

@@ -105,7 +105,7 @@ void printCenteredTitle(const char *title, char fill, int dim = 30);
 bool fileExist(const char *fname);
 void downloadWeightsifDoNotExist(const std::string& input_bin, const std::string& test_folder, const std::string& weights_url);
 void readBinaryFile(std::string fname, int size, dnnType** data_h, dnnType** data_d, int seek = 0);
-int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device = true, int limit = 10);
+int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device = true, int limit = 10, bool verbose=true);
 void printDeviceVector(int size, dnnType* vec_d, bool device = true);
 float getColor(const int c, const int x, const int max);
 void resize(int size, dnnType **data);

src/utils.cpp

@@ -83,7 +83,7 @@ void printDeviceVector(int size, dnnType* vec_d, bool device){
         delete [] vec;
 }
 
-int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device, int limit) {
+int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device, int limit, bool verbose) {
 
     dnnType *data_h, *correct_h;
     const float eps = 0.02f;
@@ -117,13 +117,15 @@ int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device, int
         delete [] correct_h;
     }
 
-    std::cout<<" | ";
-    if(diffs == 0)
-        std::cout<<COL_GREENB<<"OK";
-    else
-        std::cout<<COL_REDB<<"Wrongs: "<<diffs;
+    if(verbose){
+        std::cout<<" | ";
+        if(diffs == 0)
+            std::cout<<COL_GREENB<<"OK";
+        else
+            std::cout<<COL_REDB<<"Wrongs: "<<diffs;
 
-    std::cout<<COL_END<<" ~"<<eps<<"\n";
+        std::cout<<COL_END<<" ~"<<eps<<"\n";
+    }
     return diffs;
 }
 

tests/darknet/yolo4_320.cpp

@@ -5,7 +5,7 @@
 #include "DarknetParser.h"
 
 int main() {
-    std::string bin_path  = "yolo4";
+    std::string bin_path  = "yolo4_320";
     std::vector<std::string> input_bins = { 
         bin_path + "/layers/input.bin"
     };
@@ -15,9 +15,9 @@ int main() {
         bin_path + "/debug/layer161_out.bin"
     };
     std::string wgs_path  = bin_path + "/layers";
-    std::string cfg_path  = "../tests/darknet/cfg/yolo4.cfg";
+    std::string cfg_path  = "../tests/darknet/cfg/yolo4_320.cfg";
     std::string name_path = "../tests/darknet/names/coco.names";
-    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/d97CFzYqCPCp5Hg/download");
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/64PHAwrM6RCZbiR/download");
 
     // parse darknet network
     tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);

tests/darknet/yolo4_416.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_416";
+    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_416.cfg";
+    std::string name_path = "../tests/darknet/names/coco.names";
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/982LxTQcNQfFQc4/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;
+}

tests/darknet/yolo4_512.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_512";
+    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_512.cfg";
+    std::string name_path = "../tests/darknet/names/coco.names";
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/XN3FNXs3fnMaK5i/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;
+}

tests/darknet/yolo4_608.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_608";
+    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_608.cfg";
+    std::string name_path = "../tests/darknet/names/coco.names";
+    downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/Bg9r7kqDFJiFB4c/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;
+}

tests/test_rtinference/rtinference.cpp

@@ -1,4 +1,5 @@
 #include<iostream>
+#include<algorithm>
 #include "tkdnn.h"
 #include <stdlib.h>     /* srand, rand */
 
@@ -29,6 +30,7 @@ int main(int argc, char *argv[]) {
 
     int ret_tensorrt = 0; 
     std::cout<<"Testing with batchsize: "<<BATCH_SIZE<<"\n";
+    std::vector<double> stats;
     printCenteredTitle(" TENSORRT inference ", '=', 30); 
     float total_time = 0;
     for(int i=0; i<1200; i++) {
@@ -46,18 +48,26 @@ int main(int argc, char *argv[]) {
         netRT.infer(dim, input_d);
         TKDNN_TSTOP
         total_time+= t_ns;
+        if(i> 1)
+        stats.push_back(t_ns);
 
         // control output
-        std::cout<<"Output Buffers: "<<netRT.getBuffersN()-1<<"\n";
+        // std::cout<<"Output Buffers: "<<netRT.getBuffersN()-1<<"\n";
+        std::cout<<"Img: "<<i<<"\n";
 	for(int o=1; o<netRT.getBuffersN(); o++) {
             for(int b=1; b<BATCH_SIZE; b++) {
                 dnnType *out_d = (dnnType*) netRT.buffersRT[o];
                 dnnType *out0_d = out_d;
                 dnnType *outI_d = out_d + netRT.buffersDIM[o].tot()*b;
-                ret_tensorrt |= checkResult(netRT.buffersDIM[o].tot(), outI_d, out0_d) == 0 ? 0 : ERROR_TENSORRT;
+                ret_tensorrt |= checkResult(netRT.buffersDIM[o].tot(), outI_d, out0_d,true, 10, false) == 0 ? 0 : ERROR_TENSORRT;
             }
         }
     }
-    std::cout<<"avg: "<<total_time/1200.<<std::endl;
+    std::cout<<"Min: "<<*std::min_element(stats.begin(), stats.end())/BATCH_SIZE<<" ms\n";    
+    std::cout<<"Max: "<<*std::max_element(stats.begin(), stats.end())/BATCH_SIZE<<" ms\n";    
+    double mean =0;
+    for(int i=0; i<stats.size(); i++) mean += stats[i]; mean /= stats.size();
+    std::cout<<"Avg: "<<mean/BATCH_SIZE<<" ms\t"<<1000/(mean/BATCH_SIZE)<<" FPS\n"<<COL_END;   
     return ret_tensorrt;
 }
+