From 028696bf15efeca3acb3db8c42a96f7b9e0f55ff Mon Sep 17 00:00:00 2001
From: iovodov <b@ovdv.ru>
Date: Thu, 03 May 2018 13:33:46 +0000
Subject: [PATCH] Output improvements for detector results: When printing detector results, output was done in random order, obfuscating results for interpreting. Now: 1. Text output includes coordinates of rects in (left,right,top,bottom in pixels) along with label and score 2. Text output is sorted by rect lefts to simplify finding appropriate rects on image 3. If several class probs are > thresh for some detection, the most probable is written first and coordinates for others are not repeated 4. Rects are imprinted in image in order by their best class prob, so most probable rects are always on top and not overlayed by less probable ones 5. Most probable label for rect is always written first Also: 6. Message about low GPU memory include required amount
---
src/yolo_v2_class.hpp | 222 ++++++++++++++++++++++++++++++++++++++++++++++++------
1 files changed, 195 insertions(+), 27 deletions(-)
diff --git a/src/yolo_v2_class.hpp b/src/yolo_v2_class.hpp
index 7d0516e..00c2a5b 100644
--- a/src/yolo_v2_class.hpp
+++ b/src/yolo_v2_class.hpp
@@ -1,15 +1,4 @@
#pragma once
-#include <memory>
-#include <vector>
-#include <deque>
-#include <algorithm>
-
-#ifdef OPENCV
-#include <opencv2/opencv.hpp> // C++
-#include "opencv2/highgui/highgui_c.h" // C
-#include "opencv2/imgproc/imgproc_c.h" // C
-#endif // OPENCV
-
#ifdef YOLODLL_EXPORTS
#if defined(_MSC_VER)
#define YOLODLL_API __declspec(dllexport)
@@ -39,6 +28,17 @@
float *data; // pointer to the image data
};
+#ifdef __cplusplus
+#include <memory>
+#include <vector>
+#include <deque>
+#include <algorithm>
+
+#ifdef OPENCV
+#include <opencv2/opencv.hpp> // C++
+#include "opencv2/highgui/highgui_c.h" // C
+#include "opencv2/imgproc/imgproc_c.h" // C
+#endif // OPENCV
class Detector {
std::shared_ptr<void> detector_gpu_ptr;
@@ -61,23 +61,23 @@
YOLODLL_API std::vector<bbox_t> tracking_id(std::vector<bbox_t> cur_bbox_vec, bool const change_history = true,
int const frames_story = 10, int const max_dist = 150);
+ std::vector<bbox_t> detect_resized(image_t img, int init_w, int init_h, float thresh = 0.2, bool use_mean = false)
+ {
+ if (img.data == NULL)
+ throw std::runtime_error("Image is empty");
+ auto detection_boxes = detect(img, thresh, use_mean);
+ float wk = (float)init_w / img.w, hk = (float)init_h / img.h;
+ for (auto &i : detection_boxes) i.x *= wk, i.w *= wk, i.y *= hk, i.h *= hk;
+ return detection_boxes;
+ }
+
#ifdef OPENCV
std::vector<bbox_t> detect(cv::Mat mat, float thresh = 0.2, bool use_mean = false)
{
if(mat.data == NULL)
throw std::runtime_error("Image is empty");
auto image_ptr = mat_to_image_resize(mat);
- return detect_resized(*image_ptr, mat.size(), thresh, use_mean);
- }
-
- std::vector<bbox_t> detect_resized(image_t img, cv::Size init_size, float thresh = 0.2, bool use_mean = false)
- {
- if (img.data == NULL)
- throw std::runtime_error("Image is empty");
- auto detection_boxes = detect(img, thresh, use_mean);
- float wk = (float)init_size.width / img.w, hk = (float)init_size.height / img.h;
- for (auto &i : detection_boxes) i.x *= wk, i.w *= wk, i.y *= hk, i.h *= hk;
- return detection_boxes;
+ return detect_resized(*image_ptr, mat.cols, mat.rows, thresh, use_mean);
}
std::shared_ptr<image_t> mat_to_image_resize(cv::Mat mat) const
@@ -145,7 +145,7 @@
-#if defined(TRACK_OPTFLOW) && defined(OPENCV)
+#if defined(TRACK_OPTFLOW) && defined(OPENCV) && defined(GPU)
#include <opencv2/cudaoptflow.hpp>
#include <opencv2/cudaimgproc.hpp>
@@ -193,8 +193,7 @@
void update_cur_bbox_vec(std::vector<bbox_t> _cur_bbox_vec)
{
cur_bbox_vec = _cur_bbox_vec;
- good_bbox_vec_flags.resize(cur_bbox_vec.size());
- for (auto &i : good_bbox_vec_flags) i = true;
+ good_bbox_vec_flags = std::vector<bool>(cur_bbox_vec.size(), true);
cv::Mat prev_pts, cur_pts_flow_cpu;
for (auto &i : cur_bbox_vec) {
@@ -297,7 +296,8 @@
float moved_y = cur_key_pt.y - prev_key_pt.y;
if (abs(moved_x) < 100 && abs(moved_y) < 100 && good_bbox_vec_flags[i])
- if (err_cpu.at<float>(0, i) < flow_error && status_cpu.at<unsigned char>(0, i) != 0)
+ if (err_cpu.at<float>(0, i) < flow_error && status_cpu.at<unsigned char>(0, i) != 0 &&
+ ((float)cur_bbox_vec[i].x + moved_x) > 0 && ((float)cur_bbox_vec[i].y + moved_y) > 0)
{
cur_bbox_vec[i].x += moved_x + 0.5;
cur_bbox_vec[i].y += moved_y + 0.5;
@@ -320,6 +320,123 @@
}
};
+
+#elif defined(TRACK_OPTFLOW) && defined(OPENCV)
+
+//#include <opencv2/optflow.hpp>
+#include <opencv2/video/tracking.hpp>
+
+class Tracker_optflow {
+public:
+ const int flow_error;
+
+
+ Tracker_optflow(int win_size = 9, int max_level = 3, int iterations = 8000, int _flow_error = -1) :
+ flow_error((_flow_error > 0)? _flow_error:(win_size*4))
+ {
+ sync_PyrLKOpticalFlow = cv::SparsePyrLKOpticalFlow::create();
+ sync_PyrLKOpticalFlow->setWinSize(cv::Size(win_size, win_size)); // 9, 15, 21, 31
+ sync_PyrLKOpticalFlow->setMaxLevel(max_level); // +- 3 pt
+
+ }
+
+ // just to avoid extra allocations
+ cv::Mat dst_grey;
+ cv::Mat prev_pts_flow, cur_pts_flow;
+ cv::Mat status, err;
+
+ cv::Mat src_grey; // used in both functions
+ cv::Ptr<cv::SparsePyrLKOpticalFlow> sync_PyrLKOpticalFlow;
+
+ std::vector<bbox_t> cur_bbox_vec;
+ std::vector<bool> good_bbox_vec_flags;
+
+ void update_cur_bbox_vec(std::vector<bbox_t> _cur_bbox_vec)
+ {
+ cur_bbox_vec = _cur_bbox_vec;
+ good_bbox_vec_flags = std::vector<bool>(cur_bbox_vec.size(), true);
+ cv::Mat prev_pts, cur_pts_flow;
+
+ for (auto &i : cur_bbox_vec) {
+ float x_center = (i.x + i.w / 2.0F);
+ float y_center = (i.y + i.h / 2.0F);
+ prev_pts.push_back(cv::Point2f(x_center, y_center));
+ }
+
+ if (prev_pts.rows == 0)
+ prev_pts_flow = cv::Mat();
+ else
+ cv::transpose(prev_pts, prev_pts_flow);
+ }
+
+
+ void update_tracking_flow(cv::Mat new_src_mat, std::vector<bbox_t> _cur_bbox_vec)
+ {
+ if (new_src_mat.channels() == 3) {
+
+ update_cur_bbox_vec(_cur_bbox_vec);
+
+ cv::cvtColor(new_src_mat, src_grey, CV_BGR2GRAY, 1);
+ }
+ }
+
+
+ std::vector<bbox_t> tracking_flow(cv::Mat new_dst_mat, bool check_error = true)
+ {
+ if (sync_PyrLKOpticalFlow.empty()) {
+ std::cout << "sync_PyrLKOpticalFlow isn't initialized \n";
+ return cur_bbox_vec;
+ }
+
+ cv::cvtColor(new_dst_mat, dst_grey, CV_BGR2GRAY, 1);
+
+ if (src_grey.rows != dst_grey.rows || src_grey.cols != dst_grey.cols) {
+ src_grey = dst_grey.clone();
+ return cur_bbox_vec;
+ }
+
+ if (prev_pts_flow.cols < 1) {
+ return cur_bbox_vec;
+ }
+
+ ////sync_PyrLKOpticalFlow_gpu.sparse(src_grey_gpu, dst_grey_gpu, prev_pts_flow_gpu, cur_pts_flow_gpu, status_gpu, &err_gpu); // OpenCV 2.4.x
+ sync_PyrLKOpticalFlow->calc(src_grey, dst_grey, prev_pts_flow, cur_pts_flow, status, err); // OpenCV 3.x
+
+ dst_grey.copyTo(src_grey);
+
+ std::vector<bbox_t> result_bbox_vec;
+
+ if (err.rows == cur_bbox_vec.size() && status.rows == cur_bbox_vec.size())
+ {
+ for (size_t i = 0; i < cur_bbox_vec.size(); ++i)
+ {
+ cv::Point2f cur_key_pt = cur_pts_flow.at<cv::Point2f>(0, i);
+ cv::Point2f prev_key_pt = prev_pts_flow.at<cv::Point2f>(0, i);
+
+ float moved_x = cur_key_pt.x - prev_key_pt.x;
+ float moved_y = cur_key_pt.y - prev_key_pt.y;
+
+ if (abs(moved_x) < 100 && abs(moved_y) < 100 && good_bbox_vec_flags[i])
+ if (err.at<float>(0, i) < flow_error && status.at<unsigned char>(0, i) != 0 &&
+ ((float)cur_bbox_vec[i].x + moved_x) > 0 && ((float)cur_bbox_vec[i].y + moved_y) > 0)
+ {
+ cur_bbox_vec[i].x += moved_x + 0.5;
+ cur_bbox_vec[i].y += moved_y + 0.5;
+ result_bbox_vec.push_back(cur_bbox_vec[i]);
+ }
+ else good_bbox_vec_flags[i] = false;
+ else good_bbox_vec_flags[i] = false;
+
+ //if(!check_error && !good_bbox_vec_flags[i]) result_bbox_vec.push_back(cur_bbox_vec[i]);
+ }
+ }
+
+ prev_pts_flow = cur_pts_flow.clone();
+
+ return result_bbox_vec;
+ }
+
+};
#else
class Tracker_optflow {};
@@ -329,7 +446,7 @@
#ifdef OPENCV
-cv::Scalar obj_id_to_color(int obj_id) {
+static cv::Scalar obj_id_to_color(int obj_id) {
int const colors[6][3] = { { 1,0,1 },{ 0,0,1 },{ 0,1,1 },{ 0,1,0 },{ 1,1,0 },{ 1,0,0 } };
int const offset = obj_id * 123457 % 6;
int const color_scale = 150 + (obj_id * 123457) % 100;
@@ -471,3 +588,54 @@
}
};
#endif // OPENCV
+
+//extern "C" {
+#endif // __cplusplus
+
+/*
+ // C - wrappers
+ YOLODLL_API void create_detector(char const* cfg_filename, char const* weight_filename, int gpu_id);
+ YOLODLL_API void delete_detector();
+ YOLODLL_API bbox_t* detect_custom(image_t img, float thresh, bool use_mean, int *result_size);
+ YOLODLL_API bbox_t* detect_resized(image_t img, int init_w, int init_h, float thresh, bool use_mean, int *result_size);
+ YOLODLL_API bbox_t* detect(image_t img, int *result_size);
+ YOLODLL_API image_t load_img(char *image_filename);
+ YOLODLL_API void free_img(image_t m);
+
+#ifdef __cplusplus
+} // extern "C"
+
+static std::shared_ptr<void> c_detector_ptr;
+static std::vector<bbox_t> c_result_vec;
+
+void create_detector(char const* cfg_filename, char const* weight_filename, int gpu_id) {
+ c_detector_ptr = std::make_shared<YOLODLL_API Detector>(cfg_filename, weight_filename, gpu_id);
+}
+
+void delete_detector() { c_detector_ptr.reset(); }
+
+bbox_t* detect_custom(image_t img, float thresh, bool use_mean, int *result_size) {
+ c_result_vec = static_cast<Detector*>(c_detector_ptr.get())->detect(img, thresh, use_mean);
+ *result_size = c_result_vec.size();
+ return c_result_vec.data();
+}
+
+bbox_t* detect_resized(image_t img, int init_w, int init_h, float thresh, bool use_mean, int *result_size) {
+ c_result_vec = static_cast<Detector*>(c_detector_ptr.get())->detect_resized(img, init_w, init_h, thresh, use_mean);
+ *result_size = c_result_vec.size();
+ return c_result_vec.data();
+}
+
+bbox_t* detect(image_t img, int *result_size) {
+ return detect_custom(img, 0.24, true, result_size);
+}
+
+image_t load_img(char *image_filename) {
+ return static_cast<Detector*>(c_detector_ptr.get())->load_image(image_filename);
+}
+void free_img(image_t m) {
+ static_cast<Detector*>(c_detector_ptr.get())->free_image(m);
+}
+
+#endif // __cplusplus
+*/
--
Gitblit v1.10.0