From 1b5afb45838e603fa6780762eb8cc59246dc2d81 Mon Sep 17 00:00:00 2001
From: IlyaOvodov <b@ovdv.ru>
Date: Tue, 08 May 2018 11:09:35 +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 | 582 +++++++++++++++++++++++++++++++++++++++++++++------------
1 files changed, 455 insertions(+), 127 deletions(-)
diff --git a/src/yolo_v2_class.hpp b/src/yolo_v2_class.hpp
index d60f359..00c2a5b 100644
--- a/src/yolo_v2_class.hpp
+++ b/src/yolo_v2_class.hpp
@@ -1,20 +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
-
-#include <opencv2/cudaoptflow.hpp>
-#include <opencv2/cudaimgproc.hpp>
-#include <opencv2/cudaarithm.hpp>
-#include <opencv2/core/cuda.hpp>
-#endif // OPENCV
-
#ifdef YOLODLL_EXPORTS
#if defined(_MSC_VER)
#define YOLODLL_API __declspec(dllexport)
@@ -34,6 +18,7 @@
float prob; // confidence - probability that the object was found correctly
unsigned int obj_id; // class of object - from range [0, classes-1]
unsigned int track_id; // tracking id for video (0 - untracked, 1 - inf - tracked object)
+ unsigned int frames_counter;// counter of frames on which the object was detected
};
struct image_t {
@@ -43,12 +28,25 @@
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;
std::deque<std::vector<bbox_t>> prev_bbox_vec_deque;
+ const int cur_gpu_id;
public:
float nms = .4;
+ bool wait_stream;
YOLODLL_API Detector(std::string cfg_filename, std::string weight_filename, int gpu_id = 0);
YOLODLL_API ~Detector();
@@ -60,7 +58,18 @@
YOLODLL_API int get_net_width() const;
YOLODLL_API int get_net_height() const;
- YOLODLL_API std::vector<bbox_t> tracking(std::vector<bbox_t> cur_bbox_vec, int const frames_story = 6);
+ 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)
@@ -68,17 +77,7 @@
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
@@ -89,12 +88,13 @@
return mat_to_image(det_mat);
}
- static std::shared_ptr<image_t> mat_to_image(cv::Mat img)
+ static std::shared_ptr<image_t> mat_to_image(cv::Mat img_src)
{
+ cv::Mat img;
+ cv::cvtColor(img_src, img, cv::COLOR_RGB2BGR);
std::shared_ptr<image_t> image_ptr(new image_t, [](image_t *img) { free_image(*img); delete img; });
std::shared_ptr<IplImage> ipl_small = std::make_shared<IplImage>(img);
*image_ptr = ipl_to_image(ipl_small.get());
- rgbgr_image(*image_ptr);
return image_ptr;
}
@@ -108,15 +108,17 @@
int c = src->nChannels;
int step = src->widthStep;
image_t out = make_image_custom(w, h, c);
- int i, j, k, count = 0;;
+ int count = 0;
- for (k = 0; k < c; ++k) {
- for (i = 0; i < h; ++i) {
- for (j = 0; j < w; ++j) {
- out.data[count++] = data[i*step + j*c + k] / 255.;
+ for (int k = 0; k < c; ++k) {
+ for (int i = 0; i < h; ++i) {
+ int i_step = i*step;
+ for (int j = 0; j < w; ++j) {
+ out.data[count++] = data[i_step + j*c + k] / 255.;
}
}
}
+
return out;
}
@@ -137,122 +139,74 @@
return out;
}
- static void rgbgr_image(image_t im)
- {
- int i;
- for (i = 0; i < im.w*im.h; ++i) {
- float swap = im.data[i];
- im.data[i] = im.data[i + im.w*im.h * 2];
- im.data[i + im.w*im.h * 2] = swap;
- }
- }
-
#endif // OPENCV
};
-#if defined(TRACK_OPTFLOW) && defined(OPENCV)
+
+#if defined(TRACK_OPTFLOW) && defined(OPENCV) && defined(GPU)
+
+#include <opencv2/cudaoptflow.hpp>
+#include <opencv2/cudaimgproc.hpp>
+#include <opencv2/cudaarithm.hpp>
+#include <opencv2/core/cuda.hpp>
class Tracker_optflow {
public:
- int gpu_id;
+ const int gpu_count;
+ const int gpu_id;
+ const int flow_error;
- Tracker_optflow(int _gpu_id = 0) : gpu_id(_gpu_id)
+
+ Tracker_optflow(int _gpu_id = 0, int win_size = 9, int max_level = 3, int iterations = 8000, int _flow_error = -1) :
+ gpu_count(cv::cuda::getCudaEnabledDeviceCount()), gpu_id(std::min(_gpu_id, gpu_count-1)),
+ flow_error((_flow_error > 0)? _flow_error:(win_size*4))
{
int const old_gpu_id = cv::cuda::getDevice();
- static const int gpu_count = cv::cuda::getCudaEnabledDeviceCount();
- if (gpu_count > gpu_id)
- cv::cuda::setDevice(gpu_id);
+ cv::cuda::setDevice(gpu_id);
+
+ stream = cv::cuda::Stream();
sync_PyrLKOpticalFlow_gpu = cv::cuda::SparsePyrLKOpticalFlow::create();
- //sync_PyrLKOpticalFlow_gpu->setWinSize(cv::Size(31, 31)); //sync_PyrLKOpticalFlow_gpu.winSize = cv::Size(31, 31);
- //sync_PyrLKOpticalFlow_gpu->setWinSize(cv::Size(15, 15)); //sync_PyrLKOpticalFlow_gpu.winSize = cv::Size(15, 15);
+ sync_PyrLKOpticalFlow_gpu->setWinSize(cv::Size(win_size, win_size)); // 9, 15, 21, 31
+ sync_PyrLKOpticalFlow_gpu->setMaxLevel(max_level); // +- 3 pt
+ sync_PyrLKOpticalFlow_gpu->setNumIters(iterations); // 2000, def: 30
- sync_PyrLKOpticalFlow_gpu->setWinSize(cv::Size(21, 21));
- sync_PyrLKOpticalFlow_gpu->setMaxLevel(50); //sync_PyrLKOpticalFlow_gpu.maxLevel = 8; // +-32 points // def: 3
- sync_PyrLKOpticalFlow_gpu->setNumIters(6000); //sync_PyrLKOpticalFlow_gpu.iters = 8000; // def: 30
+ cv::cuda::setDevice(old_gpu_id);
}
// just to avoid extra allocations
cv::cuda::GpuMat src_mat_gpu;
cv::cuda::GpuMat dst_mat_gpu, dst_grey_gpu;
- cv::cuda::GpuMat tmp_grey_gpu;
cv::cuda::GpuMat prev_pts_flow_gpu, cur_pts_flow_gpu;
cv::cuda::GpuMat status_gpu, err_gpu;
cv::cuda::GpuMat src_grey_gpu; // used in both functions
cv::Ptr<cv::cuda::SparsePyrLKOpticalFlow> sync_PyrLKOpticalFlow_gpu;
+ cv::cuda::Stream stream;
- void update_tracking_flow(cv::Mat src_mat)
+ std::vector<bbox_t> cur_bbox_vec;
+ std::vector<bool> good_bbox_vec_flags;
+ cv::Mat prev_pts_flow_cpu;
+
+ void update_cur_bbox_vec(std::vector<bbox_t> _cur_bbox_vec)
{
- int const old_gpu_id = cv::cuda::getDevice();
- static const int gpu_count = cv::cuda::getCudaEnabledDeviceCount();
- if (gpu_count > gpu_id)
- cv::cuda::setDevice(gpu_id);
-
- cv::cuda::Stream stream;
-
- if (src_mat.channels() == 3) {
- if (src_mat_gpu.cols == 0) {
- src_mat_gpu = cv::cuda::GpuMat(src_mat.size(), src_mat.type());
- src_grey_gpu = cv::cuda::GpuMat(src_mat.size(), CV_8UC1);
- }
-
- src_mat_gpu.upload(src_mat, stream);
- cv::cuda::cvtColor(src_mat_gpu, src_grey_gpu, CV_BGR2GRAY, 0, stream);
- }
- cv::cuda::setDevice(old_gpu_id);
- }
-
-
- std::vector<bbox_t> tracking_flow(cv::Mat dst_mat, std::vector<bbox_t> cur_bbox_vec)
- {
- if (sync_PyrLKOpticalFlow_gpu.empty()) {
- std::cout << "sync_PyrLKOpticalFlow_gpu isn't initialized \n";
- return cur_bbox_vec;
- }
-
- int const old_gpu_id = cv::cuda::getDevice();
- static const int gpu_count = cv::cuda::getCudaEnabledDeviceCount();
- if (gpu_count > gpu_id)
- cv::cuda::setDevice(gpu_id);
-
- cv::cuda::Stream stream;
-
- if (dst_mat_gpu.cols == 0) {
- dst_mat_gpu = cv::cuda::GpuMat(dst_mat.size(), dst_mat.type());
- dst_grey_gpu = cv::cuda::GpuMat(dst_mat.size(), CV_8UC1);
- tmp_grey_gpu = cv::cuda::GpuMat(dst_mat.size(), CV_8UC1);
- }
-
-
- dst_mat_gpu.upload(dst_mat, stream);
-
- cv::cuda::cvtColor(dst_mat_gpu, dst_grey_gpu, CV_BGR2GRAY, 0, stream);
-
- if (src_grey_gpu.rows != dst_grey_gpu.rows || src_grey_gpu.cols != dst_grey_gpu.cols) {
- stream.waitForCompletion();
- src_grey_gpu = dst_grey_gpu.clone();
- cv::cuda::setDevice(old_gpu_id);
- return cur_bbox_vec;
- }
-
- cv::Mat prev_pts, prev_pts_flow_cpu, cur_pts_flow_cpu;
+ 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_cpu;
for (auto &i : cur_bbox_vec) {
- float x_center = (i.x + i.w / 2);
- float y_center = (i.y + i.h / 2);
+ 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_cpu = cv::Mat();
else
cv::transpose(prev_pts, prev_pts_flow_cpu);
-
if (prev_pts_flow_gpu.cols < prev_pts_flow_cpu.cols) {
prev_pts_flow_gpu = cv::cuda::GpuMat(prev_pts_flow_cpu.size(), prev_pts_flow_cpu.type());
cur_pts_flow_gpu = cv::cuda::GpuMat(prev_pts_flow_cpu.size(), prev_pts_flow_cpu.type());
@@ -262,16 +216,66 @@
}
prev_pts_flow_gpu.upload(cv::Mat(prev_pts_flow_cpu), stream);
+ }
- dst_grey_gpu.copyTo(tmp_grey_gpu, stream);
+ void update_tracking_flow(cv::Mat src_mat, std::vector<bbox_t> _cur_bbox_vec)
+ {
+ int const old_gpu_id = cv::cuda::getDevice();
+ if (old_gpu_id != gpu_id)
+ cv::cuda::setDevice(gpu_id);
+
+ if (src_mat.channels() == 3) {
+ if (src_mat_gpu.cols == 0) {
+ src_mat_gpu = cv::cuda::GpuMat(src_mat.size(), src_mat.type());
+ src_grey_gpu = cv::cuda::GpuMat(src_mat.size(), CV_8UC1);
+ }
+
+ update_cur_bbox_vec(_cur_bbox_vec);
+
+ //src_grey_gpu.upload(src_mat, stream); // use BGR
+ src_mat_gpu.upload(src_mat, stream);
+ cv::cuda::cvtColor(src_mat_gpu, src_grey_gpu, CV_BGR2GRAY, 1, stream);
+ }
+ if (old_gpu_id != gpu_id)
+ cv::cuda::setDevice(old_gpu_id);
+ }
+
+
+ std::vector<bbox_t> tracking_flow(cv::Mat dst_mat, bool check_error = true)
+ {
+ if (sync_PyrLKOpticalFlow_gpu.empty()) {
+ std::cout << "sync_PyrLKOpticalFlow_gpu isn't initialized \n";
+ return cur_bbox_vec;
+ }
+
+ int const old_gpu_id = cv::cuda::getDevice();
+ if(old_gpu_id != gpu_id)
+ cv::cuda::setDevice(gpu_id);
+
+ if (dst_mat_gpu.cols == 0) {
+ dst_mat_gpu = cv::cuda::GpuMat(dst_mat.size(), dst_mat.type());
+ dst_grey_gpu = cv::cuda::GpuMat(dst_mat.size(), CV_8UC1);
+ }
+
+ //dst_grey_gpu.upload(dst_mat, stream); // use BGR
+ dst_mat_gpu.upload(dst_mat, stream);
+ cv::cuda::cvtColor(dst_mat_gpu, dst_grey_gpu, CV_BGR2GRAY, 1, stream);
+
+ if (src_grey_gpu.rows != dst_grey_gpu.rows || src_grey_gpu.cols != dst_grey_gpu.cols) {
+ stream.waitForCompletion();
+ src_grey_gpu = dst_grey_gpu.clone();
+ cv::cuda::setDevice(old_gpu_id);
+ 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_gpu->calc(src_grey_gpu, dst_grey_gpu, prev_pts_flow_gpu, cur_pts_flow_gpu, status_gpu, err_gpu, stream); // OpenCV 3.x
+ cv::Mat cur_pts_flow_cpu;
cur_pts_flow_gpu.download(cur_pts_flow_cpu, stream);
- tmp_grey_gpu.copyTo(src_grey_gpu, stream);
+ dst_grey_gpu.copyTo(src_grey_gpu, stream);
cv::Mat err_cpu, status_cpu;
err_gpu.download(err_cpu, stream);
@@ -281,25 +285,153 @@
std::vector<bbox_t> result_bbox_vec;
- for (size_t i = 0; i < cur_bbox_vec.size(); ++i)
+ if (err_cpu.cols == cur_bbox_vec.size() && status_cpu.cols == cur_bbox_vec.size())
{
- cv::Point2f cur_key_pt = cur_pts_flow_cpu.at<cv::Point2f>(0, i);
- cv::Point2f prev_key_pt = prev_pts_flow_cpu.at<cv::Point2f>(0, i);
+ for (size_t i = 0; i < cur_bbox_vec.size(); ++i)
+ {
+ cv::Point2f cur_key_pt = cur_pts_flow_cpu.at<cv::Point2f>(0, i);
+ cv::Point2f prev_key_pt = prev_pts_flow_cpu.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;
+ float moved_x = cur_key_pt.x - prev_key_pt.x;
+ float moved_y = cur_key_pt.y - prev_key_pt.y;
- if (err_cpu.cols > i && status_cpu.cols > i)
- if (abs(moved_x) < 100 && abs(moved_y) < 100)
- //if (err_cpu.at<float>(0, i) < 60 && status_cpu.at<unsigned char>(0, i) != 0)
+ 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 &&
+ ((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]);
+ }
}
- cv::cuda::setDevice(old_gpu_id);
+ cur_pts_flow_gpu.swap(prev_pts_flow_gpu);
+ cur_pts_flow_cpu.copyTo(prev_pts_flow_cpu);
+
+ if (old_gpu_id != gpu_id)
+ cv::cuda::setDevice(old_gpu_id);
+
+ return result_bbox_vec;
+ }
+
+};
+
+#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;
}
@@ -309,5 +441,201 @@
class Tracker_optflow {};
-#endif // defined(TRACK_OPTFLOW) && defined(OPENCV)
+#endif // defined(TRACK_OPTFLOW) && defined(OPENCV)
+
+#ifdef OPENCV
+
+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;
+ cv::Scalar color(colors[offset][0], colors[offset][1], colors[offset][2]);
+ color *= color_scale;
+ return color;
+}
+
+class preview_boxes_t {
+ enum { frames_history = 30 }; // how long to keep the history saved
+
+ struct preview_box_track_t {
+ unsigned int track_id, obj_id, last_showed_frames_ago;
+ bool current_detection;
+ bbox_t bbox;
+ cv::Mat mat_obj, mat_resized_obj;
+ preview_box_track_t() : track_id(0), obj_id(0), last_showed_frames_ago(frames_history), current_detection(false) {}
+ };
+ std::vector<preview_box_track_t> preview_box_track_id;
+ size_t const preview_box_size, bottom_offset;
+ bool const one_off_detections;
+public:
+ preview_boxes_t(size_t _preview_box_size = 100, size_t _bottom_offset = 100, bool _one_off_detections = false) :
+ preview_box_size(_preview_box_size), bottom_offset(_bottom_offset), one_off_detections(_one_off_detections)
+ {}
+
+ void set(cv::Mat src_mat, std::vector<bbox_t> result_vec)
+ {
+ size_t const count_preview_boxes = src_mat.cols / preview_box_size;
+ if (preview_box_track_id.size() != count_preview_boxes) preview_box_track_id.resize(count_preview_boxes);
+
+ // increment frames history
+ for (auto &i : preview_box_track_id)
+ i.last_showed_frames_ago = std::min((unsigned)frames_history, i.last_showed_frames_ago + 1);
+
+ // occupy empty boxes
+ for (auto &k : result_vec) {
+ bool found = false;
+ // find the same (track_id)
+ for (auto &i : preview_box_track_id) {
+ if (i.track_id == k.track_id) {
+ if (!one_off_detections) i.last_showed_frames_ago = 0; // for tracked objects
+ found = true;
+ break;
+ }
+ }
+ if (!found) {
+ // find empty box
+ for (auto &i : preview_box_track_id) {
+ if (i.last_showed_frames_ago == frames_history) {
+ if (!one_off_detections && k.frames_counter == 0) break; // don't show if obj isn't tracked yet
+ i.track_id = k.track_id;
+ i.obj_id = k.obj_id;
+ i.bbox = k;
+ i.last_showed_frames_ago = 0;
+ break;
+ }
+ }
+ }
+ }
+
+ // draw preview box (from old or current frame)
+ for (size_t i = 0; i < preview_box_track_id.size(); ++i)
+ {
+ // get object image
+ cv::Mat dst = preview_box_track_id[i].mat_resized_obj;
+ preview_box_track_id[i].current_detection = false;
+
+ for (auto &k : result_vec) {
+ if (preview_box_track_id[i].track_id == k.track_id) {
+ if (one_off_detections && preview_box_track_id[i].last_showed_frames_ago > 0) {
+ preview_box_track_id[i].last_showed_frames_ago = frames_history; break;
+ }
+ bbox_t b = k;
+ cv::Rect r(b.x, b.y, b.w, b.h);
+ cv::Rect img_rect(cv::Point2i(0, 0), src_mat.size());
+ cv::Rect rect_roi = r & img_rect;
+ if (rect_roi.width > 1 || rect_roi.height > 1) {
+ cv::Mat roi = src_mat(rect_roi);
+ cv::resize(roi, dst, cv::Size(preview_box_size, preview_box_size), cv::INTER_NEAREST);
+ preview_box_track_id[i].mat_obj = roi.clone();
+ preview_box_track_id[i].mat_resized_obj = dst.clone();
+ preview_box_track_id[i].current_detection = true;
+ preview_box_track_id[i].bbox = k;
+ }
+ break;
+ }
+ }
+ }
+ }
+
+
+ void draw(cv::Mat draw_mat, bool show_small_boxes = false)
+ {
+ // draw preview box (from old or current frame)
+ for (size_t i = 0; i < preview_box_track_id.size(); ++i)
+ {
+ auto &prev_box = preview_box_track_id[i];
+
+ // draw object image
+ cv::Mat dst = prev_box.mat_resized_obj;
+ if (prev_box.last_showed_frames_ago < frames_history &&
+ dst.size() == cv::Size(preview_box_size, preview_box_size))
+ {
+ cv::Rect dst_rect_roi(cv::Point2i(i * preview_box_size, draw_mat.rows - bottom_offset), dst.size());
+ cv::Mat dst_roi = draw_mat(dst_rect_roi);
+ dst.copyTo(dst_roi);
+
+ cv::Scalar color = obj_id_to_color(prev_box.obj_id);
+ int thickness = (prev_box.current_detection) ? 5 : 1;
+ cv::rectangle(draw_mat, dst_rect_roi, color, thickness);
+
+ unsigned int const track_id = prev_box.track_id;
+ std::string track_id_str = (track_id > 0) ? std::to_string(track_id) : "";
+ putText(draw_mat, track_id_str, dst_rect_roi.tl() - cv::Point2i(-4, 5), cv::FONT_HERSHEY_COMPLEX_SMALL, 0.9, cv::Scalar(0, 0, 0), 2);
+
+ std::string size_str = std::to_string(prev_box.bbox.w) + "x" + std::to_string(prev_box.bbox.h);
+ putText(draw_mat, size_str, dst_rect_roi.tl() + cv::Point2i(0, 12), cv::FONT_HERSHEY_COMPLEX_SMALL, 0.8, cv::Scalar(0, 0, 0), 1);
+
+ if (!one_off_detections && prev_box.current_detection) {
+ cv::line(draw_mat, dst_rect_roi.tl() + cv::Point2i(preview_box_size, 0),
+ cv::Point2i(prev_box.bbox.x, prev_box.bbox.y + prev_box.bbox.h),
+ color);
+ }
+
+ if (one_off_detections && show_small_boxes) {
+ cv::Rect src_rect_roi(cv::Point2i(prev_box.bbox.x, prev_box.bbox.y),
+ cv::Size(prev_box.bbox.w, prev_box.bbox.h));
+ unsigned int const color_history = (255 * prev_box.last_showed_frames_ago) / frames_history;
+ color = cv::Scalar(255 - 3 * color_history, 255 - 2 * color_history, 255 - 1 * color_history);
+ if (prev_box.mat_obj.size() == src_rect_roi.size()) {
+ prev_box.mat_obj.copyTo(draw_mat(src_rect_roi));
+ }
+ cv::rectangle(draw_mat, src_rect_roi, color, thickness);
+ putText(draw_mat, track_id_str, src_rect_roi.tl() - cv::Point2i(0, 10), cv::FONT_HERSHEY_COMPLEX_SMALL, 0.8, cv::Scalar(0, 0, 0), 1);
+ }
+ }
+ }
+ }
+};
+#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