Add GPU version for yolo3detection

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

include/tkDNN/DetectionNN.h

@@ -3,7 +3,7 @@
 
 #include <iostream>
 #include <signal.h>
-#include <stdlib.h>     /* srand, rand */
+#include <stdlib.h>    
 #include <unistd.h>
 #include <mutex>
 #include "utils.h"
@@ -14,7 +14,7 @@
 
 #include "tkdnn.h"
 
-// #define OPENCV_CUDA //if OPENCV has been compiled with CUDA and contrib.
+#define OPENCV_CUDA //if OPENCV has been compiled with CUDA and contrib.
 
 namespace tk { namespace dnn {
 

include/tkDNN/Yolo3Detection.h

@@ -18,6 +18,8 @@ private:
 
     tk::dnn::Yolo* getYoloLayer(int n=0);
 
+    cv::Mat bgr_h;
+    
 public:
     Yolo3Detection() {};
     ~Yolo3Detection() {}; 

src/Yolo3Detection.cpp

@@ -62,9 +62,10 @@ void Yolo3Detection::preprocess(cv::Mat &frame)
 
     //write channels
     for(int i=0; i<netRT->input_dim.c; i++) {
-        int idx = i*imagePreproc.rows*imagePreproc.cols;
+        int size = imagePreproc.rows * imagePreproc.cols;
         int ch = netRT->input_dim.c-1 -i;
-        checkCuda( cudaMemcpy((void*)&input_d[idx], (void*)bgr[ch].data, imagePreproc.rows*imagePreproc.cols*sizeof(dnnType), cudaMemcpyDeviceToDevice));
+        bgr[ch].download(bgr_h); //TODO: don't copy back on CPU
+        checkCuda( cudaMemcpy(input_d + i*size, (float*)bgr_h.data, size*sizeof(dnnType), cudaMemcpyHostToDevice));
     }
 #else
     cv::resize(frame, frame, cv::Size(netRT->input_dim.w, netRT->input_dim.h));
@@ -97,6 +98,9 @@ void Yolo3Detection::update(cv::Mat &frame)
     //do inference
     tk::dnn::dataDim_t dim = netRT->input_dim;
 
+    // printDeviceVector(netRT->input_dim.tot()*sizeof(dnnType),input_d);
+    
+    
     printCenteredTitle(" TENSORRT inference ", '=', 30); 
     {
         dim.print();