#include "DnDetect.h"
|
|
DnDetect::DnDetect(const int gpuIndex):m_net(nullptr),m_thresh(0.5),m_hier_thresh(0.5),m_nms(0.45),names(nullptr),alphabet(nullptr)
|
{
|
cuda_set_device(gpuIndex);
|
|
char *datacfg = "cfg/coco.data";
|
char *cfgfile = "cfg/yolov3.cfg";
|
char *weightfile = "./yolov3.weights";
|
|
double loadtime = what_time_is_it_now();
|
list *options = read_data_cfg(datacfg);
|
char *name_list = option_find_str(options, "names", "data/names.list");
|
names = get_labels(name_list);
|
|
alphabet = load_alphabet();
|
m_net = load_network(cfgfile, weightfile, 0);
|
set_batch_network(m_net, 1);
|
printf("load mod use %f seconds.\n", what_time_is_it_now()-loadtime);
|
|
srand(2222222);
|
double attime;
|
float nms=.45;
|
}
|
|
DnDetect::~DnDetect()
|
{
|
|
}
|
|
std::vector<cv::Rect2f> DnDetect::detect(cv::Mat & img)
|
{
|
|
double bttime=what_time_is_it_now();
|
image im = matToImg(img);
|
image sized = letterbox_image(im, m_net->w, m_net->h);
|
layer l = m_net->layers[m_net->n-1];
|
|
float *X = sized.data;
|
//attime=what_time_is_it_now();
|
network_predict(m_net, X);
|
//printf("Predicted in %f seconds.\n", what_time_is_it_now()-attime);
|
int nboxes = 0;
|
detection *dets = get_network_boxes(m_net, im.w, im.h, 0.5f, 0.1f, 0, 1, &nboxes);
|
if (m_nms) do_nms_sort(dets, nboxes, l.classes, m_nms);
|
// draw_detections(im, dets, nboxes, m_thresh, names, alphabet, l.classes);
|
std::vector<cv::Rect2f> rects;
|
for(int i=0;i<nboxes;i++)
|
{
|
std::vector<float> vec(80);
|
memcpy(&vec[0],dets[i].prob,sizeof(float)*80);
|
int type = -1;
|
for(int j = 0; j < l.classes; ++j){
|
if(j != 0){
|
continue;
|
}
|
if (dets[i].prob[j] > m_thresh){
|
if (type < 0) {
|
type = j;
|
} else {
|
}
|
} else{
|
}
|
}
|
if(type >= 0){
|
if(type != 0){
|
continue;
|
}
|
float left = (dets[i].bbox.x-dets[i].bbox.w/2.);
|
float top = (dets[i].bbox.y-dets[i].bbox.h/2.);
|
cv::Rect2f rect(left,top,dets[i].bbox.w,dets[i].bbox.h);
|
std::cout<<"rect.x : "<<left<<" rect.y : "<<top<<" rect.w : "<<rect.width<<" rect.h : "<<rect.height<<std::endl;
|
rects.push_back(rect);
|
}
|
|
|
}
|
free_detections(dets, nboxes);
|
// show_image(im, "Video");
|
// cv::waitKey(10);
|
free_image(im);
|
free_image(sized);
|
printf("all time use %f seconds.\n", what_time_is_it_now()-bttime);
|
return rects;
|
}
|
|
image DnDetect::matToImg(cv::Mat& RefImg) {
|
CV_Assert(RefImg.depth() == CV_8U);
|
|
int h = RefImg.rows;
|
int w = RefImg.cols;
|
int channels = RefImg.channels();
|
image im = make_image(w, h, 3);
|
int count = 0;
|
switch(channels){
|
case 1:{
|
cv::MatIterator_<unsigned char> it, end;
|
for (it = RefImg.begin<unsigned char>(), end = RefImg.end<unsigned char>(); it != end; ++it){
|
im.data[count] = im.data[w*h + count] = im.data[w*h*2 + count] = (float)(*it)/255.0;
|
|
++count;
|
}
|
break;
|
}
|
case 3:{
|
cv::MatIterator_<cv::Vec3b> it, end;
|
for (it = RefImg.begin<cv::Vec3b>(), end = RefImg.end<cv::Vec3b>(); it != end; ++it){
|
im.data[count] = (float)(*it)[2]/255.0;
|
im.data[w*h + count] = (float)(*it)[1]/255.0;
|
im.data[w*h*2 + count] = (float)(*it)[0]/255.0;
|
++count;
|
}
|
break;
|
}
|
|
default:
|
printf("Channel number not supported.\n");
|
break;
|
}
|
return im;
|
}
|
