mmr/tkDNN
Archived
1
0
Fork 0
Code Issues 46 Pull Requests 12 Releases 4 Activity

Merge branch 'ceccocats:master' into master

Browse Source
This commit is contained in:
Harshvardhan Chandirasekar
2021-07-21 12:44:24 +05:30
committed by GitHub
parent 4ed247df58 89b1bb7bee
commit 548e87fc0e
32 changed files with 5428 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files
demo/demo/map.cpp
+101 -74
View File
@@ -34,10 +34,12 @@ int main(int argc, char *argv[])
const char *config_filename = "../demo/config.yaml";
const char * net = "yolo3.rt";
const char * labels_path = "../demo/COCO_val2017/all_labels.txt";
int n_batches = 1;
float confidence_thresh = 0.3;
bool show = false;
bool write_dets = false;
bool write_res_on_file = true;
bool write_coco_json = true;
bool write_coco_json = false;
int n_images = 5000;
bool verbose;
@@ -56,6 +58,12 @@ int main(int argc, char *argv[])
labels_path = argv[3];
if(argc > 4)
config_filename = argv[4];
if(argc > 5)
n_batches = atoi(argv[5]);
if(argc > 6)
confidence_thresh = atof(argv[6]);
std::cout<<"conf t: "<<confidence_thresh<<std::endl;
//check if files needed exist
if(!fileExist(config_filename))
@@ -83,9 +91,9 @@ int main(int argc, char *argv[])
}
if(write_res_on_file){
times.open("times_"+net_name+".csv");
times.open("times_"+net_name+"_"+ std::to_string(n_batches)+"_"+std::to_string(confidence_thresh)+".csv");
memory.open("memory.csv", std::ios_base::app);
memory<<net<<";";
memory<<net_name+"_"+ std::to_string(n_batches)+"_"+std::to_string(confidence_thresh)<<";";
}
// instantiate detector
@@ -121,90 +129,109 @@ int main(int argc, char *argv[])
if(show)
cv::namedWindow("detection", cv::WINDOW_NORMAL);
bool file_ok = false;
int images_done;
for (images_done=0 ; std::getline(all_labels, l_filename) && images_done < n_images ; ++images_done) {
std::cout <<COL_ORANGEB<< "Images done:\t" << images_done<< "\n"<<COL_END;
for (images_done=0 ; images_done < n_images ;) {
tk::dnn::Frame f;
f.lFilename = l_filename;
f.iFilename = l_filename;
convertFilename(f.iFilename, "labels", "images", ".txt", ".jpg");
// read frame
if(!fileExist(f.iFilename.c_str()))
FatalError("Wrong image file path.");
cv::Mat frame = cv::imread(f.iFilename.c_str(), cv::IMREAD_COLOR);
int cur_batches = 0;
std::vector<cv::Mat> batch_frames;
batch_frames.push_back(frame);
int height = frame.rows;
int width = frame.cols;
if(!frame.data)
break;
std::vector<cv::Mat> batch_dnn_input;
batch_dnn_input.push_back(frame.clone());
std::vector<tk::dnn::Frame> cur_frames;
for(;cur_batches<n_batches && images_done < n_images;cur_batches++, ++images_done){
std::getline(all_labels, l_filename);
file_ok = all_labels ? true : false ;
if (!file_ok)
break;
tk::dnn::Frame f;
f.lFilename = l_filename;
f.iFilename = l_filename;
convertFilename(f.iFilename, "labels", "images", ".txt", ".jpg");
// read frame
if(!fileExist(f.iFilename.c_str()))
FatalError("Wrong image file path.");
cv::Mat frame = cv::imread(f.iFilename.c_str(), cv::IMREAD_COLOR);
batch_frames.push_back(frame);
f.height = frame.rows;
f.width = frame.cols;
if(!frame.data)
break;
batch_dnn_input.push_back(frame.clone());
// read and save groundtruth labels
if(fileExist(f.lFilename.c_str()))
{
std::ifstream labels(f.lFilename);
for(std::string line; std::getline(labels, line); ){
std::istringstream in(line);
tk::dnn::BoundingBox b;
in >> b.cl >> b.x >> b.y >> b.w >> b.h;
b.prob = 1;
b.truthFlag = 1;
f.gt.push_back(b);
if(show)// draw rectangle for groundtruth
cv::rectangle(batch_frames[cur_batches], cv::Point((b.x-b.w/2)*f.width, (b.y-b.h/2)*f.height), cv::Point((b.x+b.w/2)*f.width,(b.y+b.h/2)*f.height), cv::Scalar(0, 255, 0), 2);
}
}
cur_frames.push_back(f);
}
if (!file_ok)
break;
//inference
detected_bbox.clear();
detNN->update(batch_dnn_input,1,write_res_on_file, &times, write_coco_json);
detNN->update(batch_dnn_input,cur_batches,write_res_on_file, &times, write_coco_json);
detNN->draw(batch_frames);
detected_bbox = detNN->detected;
if(write_coco_json)
printJsonCOCOFormat(&coco_json, f.iFilename.c_str(), detected_bbox, classes, width, height);
for(int j=0;j<cur_frames.size(); ++j){
if(write_coco_json)
printJsonCOCOFormat(&coco_json, cur_frames[j].iFilename.c_str(), detNN->batchDetected[j], classes, cur_frames[j].width, cur_frames[j].height);
std::ofstream myfile;
if(write_dets)
myfile.open ("det/"+f.lFilename.substr(f.lFilename.find("labels/") + 7));
std::ofstream myfile;
if(write_dets)
myfile.open ("det/"+cur_frames[j].lFilename.substr(cur_frames[j].lFilename.find("labels/") + 7));
// save detections labels
for(auto d:detected_bbox){
//convert detected bb in the same format as label
//<x_center>/<image_width> <y_center>/<image_width> <width>/<image_width> <height>/<image_width>
tk::dnn::BoundingBox b;
b.x = (d.x + d.w/2) / width;
b.y = (d.y + d.h/2) / height;
b.w = d.w / width;
b.h = d.h / height;
b.prob = d.prob;
b.cl = d.cl;
f.det.push_back(b);
// save detections labels
for(auto d:detNN->batchDetected[j]){
//convert detected bb in the same format as label
//<x_center>/<image_width> <y_center>/<image_width> <width>/<image_width> <height>/<image_width>
tk::dnn::BoundingBox b;
b.x = (d.x + d.w/2) / cur_frames[j].width;
b.y = (d.y + d.h/2) / cur_frames[j].height;
b.w = d.w / cur_frames[j].width;
b.h = d.h / cur_frames[j].height;
b.prob = d.prob;
b.cl = d.cl;
cur_frames[j].det.push_back(b);
if(write_dets)
myfile << d.cl << " "<< d.prob << " "<< b.x << " "<< b.y << " "<< b.w << " "<< b.h <<"\n";
if(show)// draw rectangle for detection
cv::rectangle(batch_frames[j], cv::Point(d.x, d.y), cv::Point(d.x + d.w, d.y + d.h), cv::Scalar(0, 0, 255), 2);
}
if(write_dets)
myfile << d.cl << " "<< d.prob << " "<< b.x << " "<< b.y << " "<< b.w << " "<< b.h <<"\n";
if(show)// draw rectangle for detection
cv::rectangle(batch_frames[0], cv::Point(d.x, d.y), cv::Point(d.x + d.w, d.y + d.h), cv::Scalar(0, 0, 255), 2);
}
if(write_dets)
myfile.close();
// read and save groundtruth labels
if(fileExist(f.lFilename.c_str()))
{
std::ifstream labels(l_filename);
for(std::string line; std::getline(labels, line); ){
std::istringstream in(line);
tk::dnn::BoundingBox b;
in >> b.cl >> b.x >> b.y >> b.w >> b.h;
b.prob = 1;
b.truthFlag = 1;
f.gt.push_back(b);
if(show)// draw rectangle for groundtruth
cv::rectangle(batch_frames[0], cv::Point((b.x-b.w/2)*width, (b.y-b.h/2)*height), cv::Point((b.x+b.w/2)*width,(b.y+b.h/2)*height), cv::Scalar(0, 255, 0), 2);
}
}
myfile.close();
images.push_back(f);
images.push_back(cur_frames[j]);
if(show){
cv::imshow("detection", batch_frames[0]);
cv::waitKey(0);
if(show){
cv::imshow("detection", batch_frames[j]);
cv::waitKey(0);
}
}
std::cout <<COL_ORANGEB<< "Images done:\t" << images_done<< "\tcur batch:\t"<<cur_batches<< "\n"<<COL_END;
getMemUsage(vm, rss);
vm_total += vm;
rss_total += rss;
@@ -221,11 +248,11 @@ int main(int argc, char *argv[])
std::cout << "Avg VM[MB]: " << vm_total/images_done/1024.0 << ";Avg RSS[MB]: " << rss_total/images_done/1024.0 << std::endl;
//compute mAP
double AP = tk::dnn::computeMapNIoULevels(images,classes,IoU_thresh,conf_thresh, map_points, map_step, map_levels, verbose, write_res_on_file, net_name);
double AP = tk::dnn::computeMapNIoULevels(images,classes,IoU_thresh,confidence_thresh, map_points, map_step, map_levels, verbose, write_res_on_file, net_name+"_"+ std::to_string(n_batches)+"_"+std::to_string(confidence_thresh));
std::cout<<"mAP "<<IoU_thresh<<":"<<IoU_thresh+map_step*(map_levels-1)<<" = "<<AP<<std::endl;
//compute average precision, recall and f1score
tk::dnn::computeTPFPFN(images,classes,IoU_thresh,conf_thresh, verbose, write_res_on_file, net_name);
tk::dnn::computeTPFPFN(images,classes,IoU_thresh,confidence_thresh, verbose, write_res_on_file, net_name +"_"+ std::to_string(n_batches)+"_"+std::to_string(confidence_thresh));
if(write_res_on_file){
memory<<vm_total/images_done/1024.0<<";"<<rss_total/images_done/1024.0<<"\n";
include/tkDNN/Layer.h
+4
View File
@@ -56,6 +56,10 @@ public:
int id = 0;
bool final; //if the layer is the final one
uint n_params = 0;
uint feature_map_size = 0;
long unsigned MACC = 0;
std::string getLayerName() {
layerType_t type = getLayerType();
include/tkDNN/Network.h
+1
View File
@@ -50,6 +50,7 @@ public:
bool addLayer(Layer *l);
void print();
const char *getNetworkRTName(const char *network_name);
void adjustFeatureMapSizeWithShortcuts();
cudnnDataType_t dataType;
cudnnTensorFormat_t tensorFormat;
include/tkDNN/evaluation.h
+2
View File
@@ -18,6 +18,8 @@ struct Frame
std::string iFilename;
std::vector<BoundingBox> gt;
std::vector<BoundingBox> det;
int width;
int height;
void print() const;
};
include/tkDNN/utils.h
+2 -2
View File
@@ -40,7 +40,7 @@
#define COL_PURPLEB "\033[1;35m"
#define COL_CYANB "\033[1;36m"
#define TKDNN_VERBOSE 1
#define TKDNN_VERBOSE 0
// Simple Timer
#ifdef __linux__
@@ -118,7 +118,7 @@ void printCenteredTitle(const char *title, char fill, int dim = 30);
bool fileExist(const char *fname);
void downloadWeightsifDoNotExist(const std::string& input_bin, const std::string& test_folder, const std::string& weights_url);
void readBinaryFile(std::string fname, int size, dnnType** data_h, dnnType** data_d, int seek = 0);
int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device = true, int limit = 10);
int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device = true, int limit = 10, bool verbose=true);
void printDeviceVector(int size, dnnType* vec_d, bool device = true);
float getColor(const int c, const int x, const int max);
void resize(int size, dnnType **data);
scripts/test_all_tests.sh
+5
View File
@@ -75,10 +75,15 @@ do
test_net shelfnet
test_net shelfnet_berkeley
test_net yolo4
test_net yolo4_320
test_net yolo4_320_coco2
test_net yolo4_512
test_net yolo4_608
test_net yolo4-csp
test_net yolo4x
test_net yolo4_berkeley
test_net yolo4tiny
test_net yolo4tiny_512
test_net yolo3
test_net yolo3_berkeley
test_net yolo3_coco4
scripts/test_inference.sh
+52
View File
@@ -0,0 +1,52 @@
#!/bin/bash
function test_inference {
./test_$1
./test_rtinference $1_$2.rt 1
./test_rtinference $1_$2.rt 4
}
sudo jeston_clock
# modes=( 1 ) # only FP32
# modes=( 1 2 ) # FP32 and FP16
modes=( 1 2 3 ) # FP32, FP16 and INT8
rm times_rtinference.csv
for i in "${modes[@]}"
do
rm *rt
if [ $i -eq 1 ]
then
export TKDNN_MODE=FP32
mode=fp32
echo -e "${ORANGE}Test FP32${NC}"
fi
if [ $i -eq 2 ]
then
export TKDNN_MODE=FP16
mode=fp16
echo -e "${ORANGE}Test FP16${NC}"
fi
if [ $i -eq 3 ]
then
export TKDNN_MODE=INT8
export TKDNN_CALIB_LABEL_PATH=../demo/COCO_val2017/all_labels.txt
export TKDNN_CALIB_IMG_PATH=../demo/COCO_val2017/all_images.txt
mode=int8
echo -e "${ORANGE}Test INT8${NC}"
fi
export TKDNN_BATCHSIZE=4
echo -e "${ORANGE}Batch $TKDNN_BATCHSIZE ${NC}"
test_inference yolo4_320 $mode
test_inference yolo4 $mode
test_inference yolo4_512 $mode
test_inference yolo4_608 $mode
test_inference yolo4tiny $mode
done
src/Conv2d.cpp
+5
View File
@@ -166,6 +166,11 @@ Conv2d::Conv2d( Network *net, int out_ch, int kernelH, int kernelW,
}
initCUDNN(deConv);
if(this->groups != 1)
MACC = kernelH*kernelW*output_dim.c*output_dim.w*output_dim.h;
else
MACC = input_dim.c*kernelH*kernelW*output_dim.c*output_dim.w*output_dim.h;
// allocate warkspace
if (ws_sizeInBytes!=0) {
checkCuda( cudaMalloc(&workSpace, ws_sizeInBytes) );
src/DeformConv2d.cpp
+6
View File
@@ -73,6 +73,12 @@ DeformConv2d::DeformConv2d( Network *net, int out_ch, int deformable_group, int
output_dim.c = out_ch;
initCUDNN();
if(this->deformableGroup != 1)
MACC = kernelH*kernelW*output_dim.c*output_dim.w*output_dim.h;
else
MACC = input_dim.c*kernelH*kernelW*output_dim.c*output_dim.w*output_dim.h;
//allocate data for infer result
checkCuda( cudaMalloc(&dstData, output_dim.tot()*sizeof(dnnType)) );
}
src/Layer.cpp
+2
View File
@@ -18,6 +18,8 @@ Layer::Layer(Network *net) {
if(!net->addLayer(this))
FatalError("Net reached max number of layers");
}
feature_map_size = input_dim.tot() + output_dim.tot();
}
Layer::~Layer() {
src/LayerWgs.cpp
+3
View File
@@ -19,6 +19,8 @@ LayerWgs::LayerWgs(Network *net, int inputs, int outputs,
int seek = 0;
readBinaryFile(weights_path.c_str(), inputs*outputs*kh*kw*kl, &data_h, &data_d, seek);
seek += inputs*outputs*kh*kw*kl;
n_params = seek;
this->additional_bias = additional_bias;
if(additional_bias) {
readBinaryFile(weights_path.c_str(), outputs, &bias2_h, &bias2_d, seek);
@@ -36,6 +38,7 @@ LayerWgs::LayerWgs(Network *net, int inputs, int outputs,
readBinaryFile(weights_path.c_str(), outputs, &mean_h, &mean_d, seek);
seek += outputs;
readBinaryFile(weights_path.c_str(), outputs, &variance_h, &variance_d, seek);
seek += outputs;
float eps = TKDNN_BN_MIN_EPSILON;
src/Network.cpp
+36
View File
@@ -96,6 +96,28 @@ dataDim_t Network::getOutputDim() {
return layers[num_layers-1]->output_dim;
}
void Network::adjustFeatureMapSizeWithShortcuts(){
layerType_t layer_type;
int shortcutted_idx;
for(int i=0; i<num_layers; i++) {
layer_type = layers[i]->getLayerType();
if(layer_type == LAYER_SHORTCUT){
shortcutted_idx = -1;
for(int j=0; j<num_layers; j++) {
if(static_cast<tk::dnn::Shortcut*>(layers[i])->backLayer == layers[j]){
shortcutted_idx = j;
break;
}
}
if(shortcutted_idx == -1)
FatalError("Problem when computing featuer_map_size with shortcuts");
for(int j=shortcutted_idx+1; j<i; ++j)
layers[j]->feature_map_size += layers[shortcutted_idx]->output_dim.tot();
}
}
}
void Network::print() {
printCenteredTitle(" NETWORK MODEL ", '=', 60);
@@ -106,10 +128,21 @@ void Network::print() {
std::cout.width(16); std::cout<<std::left<<"output (H*W,CH)";
std::cout<<"\n";
adjustFeatureMapSizeWithShortcuts();
long long unsigned int tot_params = 0;
long long unsigned int max_feature_map_size = 0;
long long unsigned int tot_MACC = 0;
for(int i=0; i<num_layers; i++) {
dataDim_t in = layers[i]->input_dim;
dataDim_t out = layers[i]->output_dim;
tot_params += layers[i]->n_params;
tot_MACC += layers[i]->MACC;
if(layers[i]->feature_map_size> max_feature_map_size)
max_feature_map_size = layers[i]->feature_map_size;
std::cout.width(3); std::cout<<std::right<<i;
std::cout<<" ";
std::cout.width(16); std::cout<<std::left<<layers[i]->getLayerName();
@@ -128,6 +161,9 @@ void Network::print() {
}
printCenteredTitle("", '=', 60);
std::cout<<"\n";
std::cout<<"N params: "<<tot_params<<std::endl;
std::cout<<"Max feature map size: "<<max_feature_map_size<<std::endl;
std::cout<<"N MACC: "<<tot_MACC<<std::endl<<std::endl;
printCudaMemUsage();
}
const char *Network::getNetworkRTName(const char *network_name){
src/Yolo.cpp
-1
View File
@@ -88,7 +88,6 @@ dnnType* Yolo::infer(dataDim_t &dim, dnnType* srcData) {
for (int b = 0; b < dim.n; ++b){
for(int n = 0; n < n_masks; ++n){
int index = entry_index(b, n*dim.w*dim.h, 0, classes, input_dim, output_dim);
std::cout<<"new_coords"<<new_coords<<std::endl;
if (new_coords == 1){
if (this->scaleXY != 1) scalAdd(dstData + index, 2 * dim.w*dim.h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
}
src/utils.cpp
+9 -7
View File
@@ -92,7 +92,7 @@ void printDeviceVector(int size, dnnType* vec_d, bool device){
delete [] vec;
}
int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device, int limit) {
int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device, int limit, bool verbose) {
dnnType *data_h, *correct_h;
const float eps = 0.02f;
@@ -127,13 +127,15 @@ int checkResult(int size, dnnType *data_d, dnnType *correct_d, bool device, int
delete [] correct_h;
}
std::cout<<" | ";
if(diffs == 0)
std::cout<<COL_GREENB<<"OK";
else
std::cout<<COL_REDB<<"Wrongs: "<<diffs;
if(verbose){
std::cout<<" | ";
if(diffs == 0)
std::cout<<COL_GREENB<<"OK";
else
std::cout<<COL_REDB<<"Wrongs: "<<diffs;
std::cout<<COL_END<<" ~"<<eps<<"\n";
std::cout<<COL_END<<" ~"<<eps<<"\n";
}
return diffs;
}
tests/centernet/dla34_cnet/dla34_cnet.cpp
+12
View File
@@ -479,6 +479,18 @@ int main()
//print network model
net.print();
// for(int i=0; i<net.num_layers; i++) {
// if(net.layers[i]->getLayerType() == tk::dnn::LAYER_CONV2D) {
// tk::dnn::Conv2d *c = (tk::dnn::Conv2d*) net.layers[i];
// c->releaseDevice();
// c->releaseHost(true, false);
// }
// if(net.layers[i]->dstData != nullptr) {
// cudaFree(net.layers[i]->dstData);
// net.layers[i]->dstData = nullptr;
// }
// }
//convert network to tensorRT
tk::dnn::NetworkRT netRT(&net, net.getNetworkRTName("dla34_cnet"));
tests/centernet/resnet101_cnet/resnet101_cnet.cpp
+12
View File
@@ -353,6 +353,18 @@ int main()
//print network model
net.print();
// for(int i=0; i<net.num_layers; i++) {
// if(net.layers[i]->getLayerType() == tk::dnn::LAYER_CONV2D) {
// tk::dnn::Conv2d *c = (tk::dnn::Conv2d*) net.layers[i];
// c->releaseDevice();
// c->releaseHost(true, false);
// }
// if(net.layers[i]->dstData != nullptr) {
// cudaFree(net.layers[i]->dstData);
// net.layers[i]->dstData = nullptr;
// }
// }
//convert network to tensorRT
tk::dnn::NetworkRT netRT(&net, net.getNetworkRTName("resnet101_cnet"));
tests/darknet/cfg/yolo4_320.cfg
+1156
View File
File diff suppressed because it is too large Load Diff
tests/darknet/cfg/yolo4_320_coco2.cfg
+1158
View File
File diff suppressed because it is too large Load Diff
tests/darknet/cfg/yolo4_512.cfg
+1156
View File
File diff suppressed because it is too large Load Diff
tests/darknet/cfg/yolo4_608.cfg
+1156
View File
File diff suppressed because it is too large Load Diff
tests/darknet/cfg/yolo4tiny_512.cfg
+281
View File
@@ -0,0 +1,281 @@
[net]
# Testing
#batch=1
#subdivisions=1
# Training
batch=64
subdivisions=1
width=512
height=512
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1
learning_rate=0.00261
burn_in=1000
max_batches = 500200
policy=steps
steps=400000,450000
scales=.1,.1
[convolutional]
batch_normalize=1
filters=32
size=3
stride=2
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=64
size=3
stride=2
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky
[route]
layers=-1
groups=2
group_id=1
[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky
[route]
layers = -1,-2
[convolutional]
batch_normalize=1
filters=64
size=1
stride=1
pad=1
activation=leaky
[route]
layers = -6,-1
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky
[route]
layers=-1
groups=2
group_id=1
[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky
[route]
layers = -1,-2
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[route]
layers = -6,-1
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[route]
layers=-1
groups=2
group_id=1
[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky
[route]
layers = -1,-2
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[route]
layers = -6,-1
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
##################################
[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky
[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky
[convolutional]
size=1
stride=1
pad=1
filters=255
activation=linear
[yolo]
mask = 3,4,5
anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319
classes=80
num=6
jitter=.3
scale_x_y = 1.05
cls_normalizer=1.0
iou_normalizer=0.07
iou_loss=ciou
ignore_thresh = .7
truth_thresh = 1
random=0
resize=1.5
nms_kind=greedynms
beta_nms=0.6
[route]
layers = -4
[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky
[upsample]
stride=2
[route]
layers = -1, 23
[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky
[convolutional]
size=1
stride=1
pad=1
filters=255
activation=linear
[yolo]
mask = 1,2,3
anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319
classes=80
num=6
jitter=.3
scale_x_y = 1.05
cls_normalizer=1.0
iou_normalizer=0.07
iou_loss=ciou
ignore_thresh = .7
truth_thresh = 1
random=0
resize=1.5
nms_kind=greedynms
beta_nms=0.6
tests/darknet/names/coco2.names
+2
View File
@@ -0,0 +1,2 @@
person
stop sign
tests/darknet/yolo3_512.cpp
+12
View File
@@ -23,6 +23,18 @@ int main() {
tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
net->print();
// for(int i=0; i<net->num_layers; i++) {
// if(net->layers[i]->getLayerType() == tk::dnn::LAYER_CONV2D) {
// tk::dnn::Conv2d *c = (tk::dnn::Conv2d*) net->layers[i];
// c->releaseDevice();
// c->releaseHost(true, false);
// }
// if(net->layers[i]->dstData != nullptr) {
// cudaFree(net->layers[i]->dstData);
// net->layers[i]->dstData = nullptr;
// }
// }
//convert network to tensorRT
tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
tests/darknet/yolo3tiny_512.cpp
+13
View File
@@ -22,6 +22,19 @@ int main() {
tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
net->print();
// for(int i=0; i<net->num_layers; i++) {
// if(net->layers[i]->getLayerType() == tk::dnn::LAYER_CONV2D) {
// tk::dnn::Conv2d *c = (tk::dnn::Conv2d*) net->layers[i];
// c->releaseDevice();
// c->releaseHost(true, false);
// }
// if(net->layers[i]->dstData != nullptr) {
// cudaFree(net->layers[i]->dstData);
// net->layers[i]->dstData = nullptr;
// }
// }
//convert network to tensorRT
tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
tests/darknet/yolo4.cpp
+3 -3
View File
@@ -15,9 +15,9 @@ int main() {
bin_path + "/debug/layer161_out.bin"
};
std::string wgs_path = bin_path + "/layers";
std::string cfg_path = std::string(TKDNN_PATH) + "/tests/darknet/cfg/yolo4.cfg";
std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/coco.names";
downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/d97CFzYqCPCp5Hg/download");
std::string cfg_path = "../tests/darknet/cfg/yolo4.cfg";
std::string name_path = "../tests/darknet/names/coco.names";
downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/982LxTQcNQfFQc4/download");
// parse darknet network
tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
tests/darknet/yolo4_320.cpp
+34
View File
@@ -0,0 +1,34 @@
#include<iostream>
#include<vector>
#include "tkdnn.h"
#include "test.h"
#include "DarknetParser.h"
int main() {
std::string bin_path = "yolo4_320";
std::vector<std::string> input_bins = {
bin_path + "/layers/input.bin"
};
std::vector<std::string> output_bins = {
bin_path + "/debug/layer139_out.bin",
bin_path + "/debug/layer150_out.bin",
bin_path + "/debug/layer161_out.bin"
};
std::string wgs_path = bin_path + "/layers";
std::string cfg_path = "../tests/darknet/cfg/yolo4_320.cfg";
std::string name_path = "../tests/darknet/names/coco.names";
downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/64PHAwrM6RCZbiR/download");
// parse darknet network
tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
net->print();
//convert network to tensorRT
tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
int ret = testInference(input_bins, output_bins, net, netRT);
net->releaseLayers();
delete net;
delete netRT;
return ret;
}
tests/darknet/yolo4_320_coco2.cpp
+34
View File
@@ -0,0 +1,34 @@
#include<iostream>
#include<vector>
#include "tkdnn.h"
#include "test.h"
#include "DarknetParser.h"
int main() {
std::string bin_path = "yolo4_320_coco2";
std::vector<std::string> input_bins = {
bin_path + "/layers/input.bin"
};
std::vector<std::string> output_bins = {
bin_path + "/debug/layer139_out.bin",
bin_path + "/debug/layer150_out.bin",
bin_path + "/debug/layer161_out.bin"
};
std::string wgs_path = bin_path + "/layers";
std::string cfg_path = "../tests/darknet/cfg/yolo4_320_coco2.cfg";
std::string name_path = "../tests/darknet/names/coco2.names";
downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/f3wk99iG5y7tEr8/download");
// parse darknet network
tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
net->print();
//convert network to tensorRT
tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
int ret = testInference(input_bins, output_bins, net, netRT);
net->releaseLayers();
delete net;
delete netRT;
return ret;
}
tests/darknet/yolo4_512.cpp
+47
View File
@@ -0,0 +1,47 @@
#include<iostream>
#include<vector>
#include "tkdnn.h"
#include "test.h"
#include "DarknetParser.h"
int main() {
std::string bin_path = "yolo4_512";
std::vector<std::string> input_bins = {
bin_path + "/layers/input.bin"
};
std::vector<std::string> output_bins = {
bin_path + "/debug/layer139_out.bin",
bin_path + "/debug/layer150_out.bin",
bin_path + "/debug/layer161_out.bin"
};
std::string wgs_path = bin_path + "/layers";
std::string cfg_path = "../tests/darknet/cfg/yolo4_512.cfg";
std::string name_path = "../tests/darknet/names/coco.names";
downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/fjFDqFmiSARKxFe/download");
// parse darknet network
tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
net->print();
// for(int i=0; i<net->num_layers; i++) {
// if(net->layers[i]->getLayerType() == tk::dnn::LAYER_CONV2D) {
// tk::dnn::Conv2d *c = (tk::dnn::Conv2d*) net->layers[i];
// c->releaseDevice();
// c->releaseHost(true, false);
// }
// if(net->layers[i]->dstData != nullptr) {
// cudaFree(net->layers[i]->dstData);
// net->layers[i]->dstData = nullptr;
// }
// }
//convert network to tensorRT
tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
int ret = testInference(input_bins, output_bins, net, netRT);
net->releaseLayers();
delete net;
delete netRT;
return ret;
}
tests/darknet/yolo4_608.cpp
+34
View File
@@ -0,0 +1,34 @@
#include<iostream>
#include<vector>
#include "tkdnn.h"
#include "test.h"
#include "DarknetParser.h"
int main() {
std::string bin_path = "yolo4_608";
std::vector<std::string> input_bins = {
bin_path + "/layers/input.bin"
};
std::vector<std::string> output_bins = {
bin_path + "/debug/layer139_out.bin",
bin_path + "/debug/layer150_out.bin",
bin_path + "/debug/layer161_out.bin"
};
std::string wgs_path = bin_path + "/layers";
std::string cfg_path = "../tests/darknet/cfg/yolo4_608.cfg";
std::string name_path = "../tests/darknet/names/coco.names";
downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/Bg9r7kqDFJiFB4c/download");
// parse darknet network
tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
net->print();
//convert network to tensorRT
tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
int ret = testInference(input_bins, output_bins, net, netRT);
net->releaseLayers();
delete net;
delete netRT;
return ret;
}
tests/darknet/yolo4tiny_512.cpp
+45
View File
@@ -0,0 +1,45 @@
#include<iostream>
#include<vector>
#include "tkdnn.h"
#include "test.h"
#include "DarknetParser.h"
int main() {
std::string bin_path = "yolo4tiny_512";
std::vector<std::string> input_bins = {
bin_path + "/layers/input.bin"
};
std::vector<std::string> output_bins = {
bin_path + "/debug/layer30_out.bin",
bin_path + "/debug/layer37_out.bin"
};
std::string wgs_path = bin_path + "/layers";
std::string cfg_path = std::string(TKDNN_PATH) + "/tests/darknet/cfg/yolo4tiny_512.cfg";
std::string name_path = std::string(TKDNN_PATH) + "/tests/darknet/names/coco.names";
downloadWeightsifDoNotExist(input_bins[0], bin_path, "https://cloud.hipert.unimore.it/s/qa2ws4GXg7mS5nN/download");
// parse darknet network
tk::dnn::Network *net = tk::dnn::darknetParser(cfg_path, wgs_path, name_path);
net->print();
// for(int i=0; i<net->num_layers; i++) {
// if(net->layers[i]->getLayerType() == tk::dnn::LAYER_CONV2D) {
// tk::dnn::Conv2d *c = (tk::dnn::Conv2d*) net->layers[i];
// c->releaseDevice();
// c->releaseHost(true, false);
// }
// if(net->layers[i]->dstData != nullptr) {
// cudaFree(net->layers[i]->dstData);
// net->layers[i]->dstData = nullptr;
// }
// }
//convert network to tensorRT
tk::dnn::NetworkRT *netRT = new tk::dnn::NetworkRT(net, net->getNetworkRTName(bin_path.c_str()));
int ret = testInference(input_bins, output_bins, net, netRT);
net->releaseLayers();
delete net;
delete netRT;
return ret;
}
tests/mobilenet/mobilenetv2ssd512/mobilenetv2ssd512.cpp
+13
View File
@@ -469,6 +469,19 @@ int main()
//print network model
net.print();
// for(int i=0; i<net.num_layers; i++) {
// if(net.layers[i]->getLayerType() == tk::dnn::LAYER_CONV2D) {
// tk::dnn::Conv2d *c = (tk::dnn::Conv2d*) net.layers[i];
// c->releaseDevice();
// c->releaseHost(true, false);
// }
// if(net.layers[i]->dstData != nullptr) {
// cudaFree(net.layers[i]->dstData);
// net.layers[i]->dstData = nullptr;
// }
// }
// convert network to tensorRT
tk::dnn::NetworkRT netRT(&net, net.getNetworkRTName("mobilenetv2ssd512"));
tests/test_rtinference/rtinference.cpp
+32 -3
View File
@@ -1,4 +1,5 @@
#include<iostream>
#include<algorithm>
#include "tkdnn.h"
#include <stdlib.h> /* srand, rand */
@@ -17,6 +18,8 @@ int main(int argc, char *argv[]) {
//convert network to tensorRT
tk::dnn::NetworkRT netRT(NULL, argv[1]);
tk::dnn::dataDim_t idim = netRT.input_dim;
tk::dnn::dataDim_t odim = netRT.output_dim;
@@ -29,6 +32,7 @@ int main(int argc, char *argv[]) {
int ret_tensorrt = 0;
std::cout<<"Testing with batchsize: "<<BATCH_SIZE<<"\n";
std::vector<double> stats;
printCenteredTitle(" TENSORRT inference ", '=', 30);
float total_time = 0;
for(int i=0; i<64; i++) {
@@ -46,18 +50,43 @@ int main(int argc, char *argv[]) {
netRT.infer(dim, input_d);
TKDNN_TSTOP
total_time+= t_ns;
if(i> 1)
stats.push_back(t_ns);
// control output
std::cout<<"Output Buffers: "<<netRT.getBuffersN()-1<<"\n";
// std::cout<<"Output Buffers: "<<netRT.getBuffersN()-1<<"\n";
std::cout<<"Img: "<<i<<"\n";
for(int o=1; o<netRT.getBuffersN(); o++) {
for(int b=1; b<BATCH_SIZE; b++) {
dnnType *out_d = (dnnType*) netRT.buffersRT[o];
dnnType *out0_d = out_d;
dnnType *outI_d = out_d + netRT.buffersDIM[o].tot()*b;
ret_tensorrt |= checkResult(netRT.buffersDIM[o].tot(), outI_d, out0_d) == 0 ? 0 : ERROR_TENSORRT;
ret_tensorrt |= checkResult(netRT.buffersDIM[o].tot(), outI_d, out0_d,true, 10, false) == 0 ? 0 : ERROR_TENSORRT;
}
}
}
std::cout<<"avg: "<<total_time/64.<<std::endl;
double min = *std::min_element(stats.begin(), stats.end())/BATCH_SIZE;
double max = *std::max_element(stats.begin(), stats.end())/BATCH_SIZE;
double mean =0;
for(int i=0; i<stats.size(); i++) mean += stats[i]; mean /= stats.size();
mean /=BATCH_SIZE;
std::cout<<"Min: "<<min<<" ms\n";
std::cout<<"Max: "<<max<<" ms\n";
std::cout<<"Avg: "<<mean<<" ms\t"<<1000/(mean)<<" FPS\n"<<COL_END;
std::ofstream times;
times.open("times_rtinference.csv", std::ios_base::app);
std::string net_name;
removePathAndExtension(argv[1], net_name);
times << net_name<< "_" << BATCH_SIZE << ";" << mean << ";" << min << ";" << max << ";" << 1000./mean << "\n";
times.close();
return ret_tensorrt;
}