From d6cb0fcabcece079c6a5b511159127f585897dba Mon Sep 17 00:00:00 2001
From: Tino Hager <tino.hager@nager.at>
Date: Wed, 27 Jun 2018 21:11:33 +0000
Subject: [PATCH] optimize max object definition

---
 src/yolo_v2_class.cpp |   78 ++++++++++++++++++++++++++++++---------
 1 files changed, 60 insertions(+), 18 deletions(-)

diff --git a/src/yolo_v2_class.cpp b/src/yolo_v2_class.cpp
index 076bab8..aad5876 100644
--- a/src/yolo_v2_class.cpp
+++ b/src/yolo_v2_class.cpp
@@ -22,6 +22,49 @@
 
 #define FRAMES 3
 
+static Detector* detector;
+//static std::unique_ptr<Detector> detector;
+
+int init(const char *configurationFilename, const char *weightsFilename, int gpu) {
+    std::string configurationFilenameString;
+    configurationFilenameString = configurationFilename;
+    std::string weightsFilenameString;
+    weightsFilenameString = weightsFilename;
+
+    detector = new Detector(configurationFilenameString, weightsFilenameString, gpu);
+    return 1;
+}
+
+int detect_image(const char *filename, bbox_t_container &container) {
+    std::string filenameString;
+    filenameString = filename;
+
+    std::vector<bbox_t> detection = detector->detect(filenameString);
+    for (size_t i = 0; i < detection.size() && i < C_SHARP_MAX_OBJECTS; ++i)
+        container.candidates[i] = detection[i];
+    return detection.size();
+}
+
+int detect_mat(const uint8_t* data, const size_t data_length, bbox_t_container &container) {
+#ifdef OPENCV
+    std::vector<char> vdata(data, data + data_length);
+    cv::Mat image = imdecode(cv::Mat(vdata), 1);
+
+    std::vector<bbox_t> detection = detector->detect(image);
+    for (size_t i = 0; i < detection.size() && i < C_SHARP_MAX_OBJECTS; ++i)
+        container.candidates[i] = detection[i];
+    return detection.size();
+#else
+    return -1;
+#endif	// OPENCV
+}
+
+int dispose() {
+    detector->~Detector();
+    //detector.reset();
+    return 1;
+}
+
 #ifdef GPU
 void check_cuda(cudaError_t status) {
 	if (status != cudaSuccess) {
@@ -32,8 +75,6 @@
 #endif
 
 struct detector_gpu_t {
-	float **probs;
-	box *boxes;
 	network net;
 	image images[FRAMES];
 	float *avg;
@@ -71,6 +112,7 @@
 	}
 	set_batch_network(&net, 1);
 	net.gpu_index = cur_gpu_id;
+	fuse_conv_batchnorm(net);
 
 	layer l = net.layers[net.n - 1];
 	int j;
@@ -79,10 +121,6 @@
 	for (j = 0; j < FRAMES; ++j) detector_gpu.predictions[j] = (float *)calloc(l.outputs, sizeof(float));
 	for (j = 0; j < FRAMES; ++j) detector_gpu.images[j] = make_image(1, 1, 3);
 
-	detector_gpu.boxes = (box *)calloc(l.w*l.h*l.n, sizeof(box));
-	detector_gpu.probs = (float **)calloc(l.w*l.h*l.n, sizeof(float *));
-	for (j = 0; j < l.w*l.h*l.n; ++j) detector_gpu.probs[j] = (float *)calloc(l.classes, sizeof(float));
-
 	detector_gpu.track_id = (unsigned int *)calloc(l.classes, sizeof(unsigned int));
 	for (j = 0; j < l.classes; ++j) detector_gpu.track_id[j] = 1;
 
@@ -103,14 +141,9 @@
 	for (int j = 0; j < FRAMES; ++j) free(detector_gpu.predictions[j]);
 	for (int j = 0; j < FRAMES; ++j) if(detector_gpu.images[j].data) free(detector_gpu.images[j].data);
 
-	for (int j = 0; j < l.w*l.h*l.n; ++j) free(detector_gpu.probs[j]);
-	free(detector_gpu.boxes);
-	free(detector_gpu.probs);
-
 	int old_gpu_index;
 #ifdef GPU
 	cudaGetDevice(&old_gpu_index);
-	//cudaSetDevice(detector_gpu.net.gpu_index);
 	cuda_set_device(detector_gpu.net.gpu_index);
 #endif
 
@@ -129,6 +162,10 @@
 	detector_gpu_t &detector_gpu = *static_cast<detector_gpu_t *>(detector_gpu_ptr.get());
 	return detector_gpu.net.h;
 }
+YOLODLL_API int Detector::get_net_color_depth() const {
+	detector_gpu_t &detector_gpu = *static_cast<detector_gpu_t *>(detector_gpu_ptr.get());
+	return detector_gpu.net.c;
+}
 
 
 YOLODLL_API std::vector<bbox_t> Detector::detect(std::string image_filename, float thresh, bool use_mean)
@@ -225,17 +262,21 @@
 		l.output = detector_gpu.avg;
 		detector_gpu.demo_index = (detector_gpu.demo_index + 1) % FRAMES;
 	}
+	//get_region_boxes(l, 1, 1, thresh, detector_gpu.probs, detector_gpu.boxes, 0, 0);
+	//if (nms) do_nms_sort(detector_gpu.boxes, detector_gpu.probs, l.w*l.h*l.n, l.classes, nms);
 
-	get_region_boxes(l, 1, 1, thresh, detector_gpu.probs, detector_gpu.boxes, 0, 0);
-	if (nms) do_nms_sort(detector_gpu.boxes, detector_gpu.probs, l.w*l.h*l.n, l.classes, nms);
-	//draw_detections(im, l.w*l.h*l.n, thresh, boxes, probs, names, alphabet, l.classes);
+	int nboxes = 0;
+	int letterbox = 0;
+	float hier_thresh = 0.5;
+	detection *dets = get_network_boxes(&net, im.w, im.h, thresh, hier_thresh, 0, 1, &nboxes, letterbox);
+	if (nms) do_nms_sort(dets, nboxes, l.classes, nms);
 
 	std::vector<bbox_t> bbox_vec;
 
-	for (size_t i = 0; i < (l.w*l.h*l.n); ++i) {
-		box b = detector_gpu.boxes[i];
-		int const obj_id = max_index(detector_gpu.probs[i], l.classes);
-		float const prob = detector_gpu.probs[i][obj_id];
+	for (size_t i = 0; i < nboxes; ++i) {
+		box b = dets[i].bbox;
+		int const obj_id = max_index(dets[i].prob, l.classes);
+		float const prob = dets[i].prob[obj_id];
 		
 		if (prob > thresh) 
 		{
@@ -252,6 +293,7 @@
 		}
 	}
 
+	free_detections(dets, nboxes);
 	if(sized.data)
 		free(sized.data);
 

--
Gitblit v1.10.0