From 54f83e153549dd1a63bcc8fa5e55fb171621a989 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Wed, 17 Jan 2018 18:05:07 +0000
Subject: [PATCH] Some fixes
---
src/yolo_v2_class.cpp | 76 +++++++++++++++++++++++++++++---------
1 files changed, 58 insertions(+), 18 deletions(-)
diff --git a/src/yolo_v2_class.cpp b/src/yolo_v2_class.cpp
index 31f623c..518ea65 100644
--- a/src/yolo_v2_class.cpp
+++ b/src/yolo_v2_class.cpp
@@ -29,11 +29,14 @@
image images[FRAMES];
float *avg;
float *predictions[FRAMES];
+ int demo_index;
+ unsigned int *track_id;
};
-YOLODLL_API Detector::Detector(std::string cfg_filename, std::string weight_filename, int gpu_id)
+YOLODLL_API Detector::Detector(std::string cfg_filename, std::string weight_filename, int gpu_id) : cur_gpu_id(gpu_id)
{
+ wait_stream = 0;
int old_gpu_index;
#ifdef GPU
cudaGetDevice(&old_gpu_index);
@@ -52,7 +55,7 @@
char *cfgfile = const_cast<char *>(cfg_filename.data());
char *weightfile = const_cast<char *>(weight_filename.data());
- net = parse_network_cfg(cfgfile);
+ net = parse_network_cfg_custom(cfgfile, 1);
if (weightfile) {
load_weights(&net, weightfile);
}
@@ -70,6 +73,9 @@
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;
+
#ifdef GPU
cudaSetDevice(old_gpu_index);
#endif
@@ -81,6 +87,8 @@
detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get());
layer l = detector_gpu.net.layers[detector_gpu.net.n - 1];
+ free(detector_gpu.track_id);
+
free(detector_gpu.avg);
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);
@@ -102,12 +110,21 @@
#endif
}
+YOLODLL_API int Detector::get_net_width() const {
+ detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get());
+ return detector_gpu.net.w;
+}
+YOLODLL_API int Detector::get_net_height() const {
+ detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get());
+ return detector_gpu.net.h;
+}
-YOLODLL_API std::vector<bbox_t> Detector::detect(std::string image_filename, float thresh)
+
+YOLODLL_API std::vector<bbox_t> Detector::detect(std::string image_filename, float thresh, bool use_mean)
{
std::shared_ptr<image_t> image_ptr(new image_t, [](image_t *img) { if (img->data) free(img->data); delete img; });
*image_ptr = load_image(image_filename);
- return detect(*image_ptr, thresh);
+ return detect(*image_ptr, thresh, use_mean);
}
static image load_image_stb(char *filename, int channels)
@@ -154,15 +171,17 @@
}
}
-YOLODLL_API std::vector<bbox_t> Detector::detect(image_t img, float thresh)
+YOLODLL_API std::vector<bbox_t> Detector::detect(image_t img, float thresh, bool use_mean)
{
-
detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get());
network &net = detector_gpu.net;
int old_gpu_index;
#ifdef GPU
cudaGetDevice(&old_gpu_index);
- cudaSetDevice(net.gpu_index);
+ if(cur_gpu_id != old_gpu_index)
+ cudaSetDevice(net.gpu_index);
+
+ net.wait_stream = wait_stream; // 1 - wait CUDA-stream, 0 - not to wait
#endif
//std::cout << "net.gpu_index = " << net.gpu_index << std::endl;
@@ -174,12 +193,27 @@
im.h = img.h;
im.w = img.w;
- image sized = resize_image(im, net.w, net.h);
+ image sized;
+
+ if (net.w == im.w && net.h == im.h) {
+ sized = make_image(im.w, im.h, im.c);
+ memcpy(sized.data, im.data, im.w*im.h*im.c * sizeof(float));
+ }
+ else
+ sized = resize_image(im, net.w, net.h);
+
layer l = net.layers[net.n - 1];
float *X = sized.data;
- network_predict(net, X);
+ float *prediction = network_predict(net, X);
+
+ if (use_mean) {
+ memcpy(detector_gpu.predictions[detector_gpu.demo_index], prediction, l.outputs * sizeof(float));
+ mean_arrays(detector_gpu.predictions, FRAMES, l.outputs, detector_gpu.avg);
+ 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);
@@ -211,7 +245,8 @@
free(sized.data);
#ifdef GPU
- cudaSetDevice(old_gpu_index);
+ if (cur_gpu_id != old_gpu_index)
+ cudaSetDevice(old_gpu_index);
#endif
return bbox_vec;
@@ -219,16 +254,15 @@
YOLODLL_API std::vector<bbox_t> Detector::tracking(std::vector<bbox_t> cur_bbox_vec, int const frames_story)
{
+ detector_gpu_t &det_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get());
+
bool prev_track_id_present = false;
for (auto &i : prev_bbox_vec_deque)
if (i.size() > 0) prev_track_id_present = true;
- static unsigned int track_id = 1;
-
if (!prev_track_id_present) {
- //track_id = 1;
for (size_t i = 0; i < cur_bbox_vec.size(); ++i)
- cur_bbox_vec[i].track_id = track_id++;
+ cur_bbox_vec[i].track_id = det_gpu.track_id[cur_bbox_vec[i].obj_id]++;
prev_bbox_vec_deque.push_front(cur_bbox_vec);
if (prev_bbox_vec_deque.size() > frames_story) prev_bbox_vec_deque.pop_back();
return cur_bbox_vec;
@@ -242,7 +276,9 @@
for (size_t m = 0; m < cur_bbox_vec.size(); ++m) {
bbox_t const& k = cur_bbox_vec[m];
if (i.obj_id == k.obj_id) {
- unsigned int cur_dist = sqrt(((float)i.x - k.x)*((float)i.x - k.x) + ((float)i.y - k.y)*((float)i.y - k.y));
+ float center_x_diff = (float)(i.x + i.w/2) - (float)(k.x + k.w/2);
+ float center_y_diff = (float)(i.y + i.h/2) - (float)(k.y + k.h/2);
+ unsigned int cur_dist = sqrt(center_x_diff*center_x_diff + center_y_diff*center_y_diff);
if (cur_dist < 100 && (k.track_id == 0 || dist_vec[m] > cur_dist)) {
dist_vec[m] = cur_dist;
cur_index = m;
@@ -250,16 +286,20 @@
}
}
- bool track_id_absent = !std::any_of(cur_bbox_vec.begin(), cur_bbox_vec.end(), [&](bbox_t const& b) { return b.track_id == i.track_id; });
+ bool track_id_absent = !std::any_of(cur_bbox_vec.begin(), cur_bbox_vec.end(),
+ [&i](bbox_t const& b) { return b.track_id == i.track_id && b.obj_id == i.obj_id; });
- if (cur_index >= 0 && track_id_absent)
+ if (cur_index >= 0 && track_id_absent){
cur_bbox_vec[cur_index].track_id = i.track_id;
+ cur_bbox_vec[cur_index].w = (cur_bbox_vec[cur_index].w + i.w) / 2;
+ cur_bbox_vec[cur_index].h = (cur_bbox_vec[cur_index].h + i.h) / 2;
+ }
}
}
for (size_t i = 0; i < cur_bbox_vec.size(); ++i)
if (cur_bbox_vec[i].track_id == 0)
- cur_bbox_vec[i].track_id = track_id++;
+ cur_bbox_vec[i].track_id = det_gpu.track_id[cur_bbox_vec[i].obj_id]++;
prev_bbox_vec_deque.push_front(cur_bbox_vec);
if (prev_bbox_vec_deque.size() > frames_story) prev_bbox_vec_deque.pop_back();
--
Gitblit v1.10.0