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

Mnist works at the moment with trt8,others like yolo4tiny and mobilenet generate the engine files but crash after throwing nvifer1::CudaRuntimeError and when demo is being run ,it doesnt deserialize properly and crashes

Browse Source
This commit is contained in:
perseusdg
2021-08-29 03:18:58 +05:30
parent 3d8b1ac494
commit 2ffe07057e
12 changed files with 255 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CMakeLists.txt
+2 -1
View File
@@ -33,12 +33,13 @@ SET(CUDA_SEPARABLE_COMPILATION ON)
#set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -arch=sm_30 --compiler-options '-fPIC'")
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32)
find_package(CUDNN REQUIRED)
include_directories(${CUDNN_INCLUDE_DIR})
# compile
file(GLOB tkdnn_CUSRC "src/kernels/*.cu" "src/sorting.cu")
file(GLOB tkdnn_CUSRC "src/kernels/*.cu" "src/sorting.cu" )
cuda_include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include ${CUDA_INCLUDE_DIRS} ${CUDNN_INCLUDE_DIRS})
cuda_add_library(kernels SHARED ${tkdnn_CUSRC})
target_link_libraries(kernels ${CUDA_CUBLAS_LIBRARIES})
demo/demo/demo.cpp
+33 -16
View File
@@ -23,6 +23,18 @@ int main(int argc, char *argv[]) {
std::string net = "yolo4tiny_fp32.rt";
#ifdef __linux__
std::string cfgPath = "../tests/darknet/cfg/yolo4tiny.cfg";
#elif _WIN32
std::string cfgPath = "..\\tests\\darknet\\cfg\\yolo4tiny.cfg";
#endif
#ifdef __linux__
std::string namePath = "../tests/darknet/names/coco.names";
#elif _WIN32
std::string namePath = "..\\tests\\darknet\\names\\coco.names";
#endif
if(argc > 1)
net = argv[1];
#ifdef __linux__
@@ -31,23 +43,28 @@ int main(int argc, char *argv[]) {
std::string input = "..\\..\\..\\demo\\yolo_test.mp4";
#endif
if(argc > 2)
input = argv[2];
char ntype = 'y';
cfgPath = argv[2];
if(argc > 3)
ntype = argv[3][0];
int n_classes = 80;
namePath = argv[3];
if(argc > 4)
n_classes = atoi(argv[4]);
int n_batch = 1;
input = argv[4];
char ntype = 'y';
if(argc > 5)
n_batch = atoi(argv[5]);
bool show = true;
ntype = argv[5][0];
int n_classes = 80;
if(argc > 6)
show = atoi(argv[6]);
float conf_thresh=0.3;
n_classes = atoi(argv[6]);
int n_batch = 1;
if(argc > 7)
conf_thresh = atof(argv[7]);
n_batch = atoi(argv[7]);
bool show = true;
if(argc > 8)
show = atoi(argv[8]);
float conf_thresh=0.3;
if(argc > 9)
conf_thresh = atof(argv[9]);
if(n_batch < 1 || n_batch > 64)
FatalError("Batch dim not supported");
@@ -56,8 +73,8 @@ int main(int argc, char *argv[]) {
SAVE_RESULT = true;
tk::dnn::Yolo3Detection yolo;
tk::dnn::CenternetDetection cnet;
tk::dnn::MobilenetDetection mbnet;
//tk::dnn::CenternetDetection cnet;
//tk::dnn::MobilenetDetection mbnet;
tk::dnn::DetectionNN *detNN;
@@ -67,17 +84,17 @@ int main(int argc, char *argv[]) {
detNN = &yolo;
break;
case 'c':
detNN = &cnet;
//detNN = &cnet;
break;
case 'm':
detNN = &mbnet;
//detNN = &mbnet;
n_classes++;
break;
default:
FatalError("Network type not allowed (3rd parameter)\n");
}
detNN->init(net, n_classes, n_batch, conf_thresh);
detNN->init(net,cfgPath,namePath,n_classes,n_batch,conf_thresh);
gRun = true;
include/tkDNN/DarknetParser.h
+3
View File
@@ -47,5 +47,8 @@ namespace tk { namespace dnn {
std::vector<tk::dnn::Layer*> &netLayers, const std::vector<std::string>& names);
std::vector<std::string> darknetReadNames(const std::string& names_file);
tk::dnn::Network* darknetParser(const std::string& cfg_file, const std::string& wgs_path, const std::string& names_file);
void loadYoloInfo(const std::string &cfg_file,int lineNo,std::vector<float> &mask,std::vector<float> &anchors,int &num,int &classes,float &nms_thresh,int &nms_kind,int &coords);
void loadYoloInitInfo(int &channels,int &width,int &height,const std::string &cfg_file);
std::vector<int> noYolosLine(const std::string &cfg_file);
}}
include/tkDNN/DetectionNN.h
+1 -1
View File
@@ -87,7 +87,7 @@ class DetectionNN {
* @param n_batches maximum number of batches to use in inference
* @return true if everything is correct, false otherwise.
*/
virtual bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3) = 0;
virtual bool init(const std::string& tensor_path,const std::string& cfg_path,const std::string& name_path, const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3) = 0;
/**
* This method performs the whole detection of the NN.
include/tkDNN/NetworkRT.h
+1 -1
View File
@@ -7,6 +7,7 @@
#include "Layer.h"
#include "NvInfer.h"
#include <memory>
#include <tkDNN/kernels.h>
namespace tk { namespace dnn {
@@ -52,7 +53,6 @@ public:
class NetworkRT {
public:
include/tkDNN/Yolo3Detection.h
+4 -3
View File
@@ -4,9 +4,9 @@
#include "opencv2/opencv.hpp"
#include "DetectionNN.h"
#include "DarknetParser.h"
namespace tk { namespace dnn {
namespace tk { namespace dnn {
class Yolo3Detection : public DetectionNN
{
private:
@@ -19,12 +19,13 @@ private:
tk::dnn::Yolo* getYoloLayer(int n=0);
cv::Mat bgr_h;
std::vector<int> noYolos;
public:
Yolo3Detection() {};
~Yolo3Detection() {};
bool init(const std::string& tensor_path, const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3);
bool init(const std::string& tensor_path,const std::string& cfg_path,const std::string& name_path,const int n_classes=80, const int n_batches=1, const float conf_thresh=0.3);
void preprocess(cv::Mat &frame, const int bi=0);
void postprocess(const int bi=0,const bool mAP=false);
};
include/tkDNN/pluginsRT/ReorgRT.h
+1 -1
View File
@@ -41,7 +41,7 @@ public:
virtual int enqueue(int batchSize, const void*const * inputs, void* const* outputs, void* workspace, cudaStream_t stream) NOEXCEPT override {
reorgForward((dnnType*)reinterpret_cast<const dnnType*>(inputs[0]),
reorgForward((dnnType*)reinterpret_cast<const dnnType*>(inputs[0]),
reinterpret_cast<dnnType*>(outputs[0]),
batchSize, c, h, w, stride, stream);
return 0;
include/tkDNN/pluginsRT/YoloRT.h
+1
View File
@@ -31,6 +31,7 @@ public:
classes = readBUF<int>(buf);
num = readBUF<int>(buf);
n_masks = readBUF<int>(buf);
std::cout<<n_masks<<std::endl;
scaleXY = readBUF<float>(buf);
nms_thresh = readBUF<float>(buf);
nms_kind = readBUF<int>(buf);
src/DarknetParser.cpp
+139 -2
View File
@@ -32,6 +32,16 @@ namespace tk { namespace dnn {
return values;
}
std::vector<float> fromStringToFloatVec(const std::string& line, const char delimiter){
std::stringstream linestream(line);
std::string value;
std::vector<float> values;
while(getline(linestream,value,delimiter))
values.push_back(std::stof(value));
return values;
}
bool darknetParseFields(const std::string& line, darknetFields_t& fields){
std::string name,value;
@@ -268,7 +278,134 @@ namespace tk { namespace dnn {
}
return net;
}
std::vector<int> noYolosLine(const std::string &cfg_file){
std::ifstream if_cfg(cfg_file);
if(!if_cfg.is_open())
FatalError("cloud not open cfg file: " + cfg_file);
std::string line;
std::vector<int> lineNo;
int count = 0;
while(std::getline(if_cfg,line)){
std::size_t found = line.find("#");
if ( found != std::string::npos ) {
line = line.substr(0, found);
}
// skip empty lines
if(line.empty())
continue;
if(line == "[yolo]"){
lineNo.push_back(count);
}
count++;
}
return lineNo;
}
void loadYoloInfo(const std::string &cfg_file,int lineNo,std::vector<float> &mask,std::vector<float> &anchors,int &num,int &classes,float &nms_thresh,int &nms_kind,int &coords){
std::vector<float> maskTemp,anchorsTemp;
int classesTemp,numTemp,nmsKindTemp;
int new_coordsTemp=0;
float nmsThreshTemp=0.45;
std::ifstream if_cfg(cfg_file);
if(!if_cfg.is_open())
FatalError("cloud not open cfg file: " + cfg_file);
std::string line;
int count = 0;
while(std::getline(if_cfg,line)){
std::string name,value;
std::size_t found = line.find("#");
if ( found != std::string::npos ) {
line = line.substr(0, found);
}
// skip empty lines
if(line.empty())
continue;
if(count > lineNo && count <=20){
divideNameAndValue(line,name,value);
if(name == "mask "){
maskTemp = fromStringToFloatVec(value,',');
}
if(name == "anchors "){
anchorsTemp = fromStringToFloatVec(value,',');
}
if(name == "classes"){
classesTemp = std::stoi(value);
}
if(name == "num"){
numTemp = std::stoi(value);
}
if(name == "nms_kind"){
if(value == "greedynms"){
nmsKindTemp = 0;
}else if(value == "diounms"){
nmsKindTemp=1;
}
else{
std::cout<<"NMS NOT SUPPORTED DEFAULTING TO GREEDYNMS"<<std::endl;
nmsKindTemp=0;
}
}
if(name == "new_coords"){
new_coordsTemp = std::stoi(value);
}
if(name == "beta_nms"){
nmsThreshTemp = std::stof(value);
}
count++;
}
}
mask = maskTemp;
anchors = anchorsTemp;
num = numTemp;
nms_kind = nmsKindTemp;
nms_thresh = nmsThreshTemp;
coords = new_coordsTemp;
classes = classesTemp;
}
void loadYoloInitInfo(int &channels,int &width,int &height,const std::string &cfg_file){
std::ifstream if_cfg(cfg_file);
if(!if_cfg.is_open())
FatalError("cloud not open cfg file: " + cfg_file);
std::string line;
int count = 0;
while(std::getline(if_cfg,line)){
if(count == 7){
std::string name,value;
divideNameAndValue(line,name,value);
if(name == "width"){
width = std::stoi(value);
}
}
if(count == 8){
std::string name,value;
divideNameAndValue(line,name,value);
if(name == "height"){
height = std::stoi(value);
}
}
if(count == 9){
std::string name,value;
divideNameAndValue(line,name,value);
if(name == "channels"){
channels = std::stoi(value);
break;
}
else{
std::cerr<<"EXITING PROGRAM DUE TO INSUFFICENT DATA FROM CFG"<<std::endl;
break;
}
}
count++;
}
}
}}
src/NetworkRT.cpp
+34
View File
@@ -11,8 +11,42 @@
#include "NetworkRT.h"
#include "Int8Calibrator.h"
using namespace nvinfer1;
PluginFieldCollection tk::dnn::ActivationLeakyRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::ActivationReLUCeilingPluginCreator::mFC{};
PluginFieldCollection tk::dnn::ActivationMishRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::ActivationLogisticRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::DeformableConvRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::RegionRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::ReorgRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::UpsampleRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::ShortcutRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::ReshapeRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::MaxPoolFixedSizeRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::ResizeLayerRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::YoloRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::RouteRTPluginCreator::mFC{};
PluginFieldCollection tk::dnn::FlattenConcatRTPluginCreator::mFC{};
std::vector<PluginField> tk::dnn::ActivationLeakyRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::ActivationReLUCeilingPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::ActivationMishRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::ActivationLogisticRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::DeformableConvRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::RegionRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::ReorgRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::UpsampleRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::ShortcutRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::ReshapeRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::MaxPoolFixedSizeRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::ResizeLayerRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::YoloRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::RouteRTPluginCreator::mPluginAttributes;
std::vector<PluginField> tk::dnn::FlattenConcatRTPluginCreator::mPluginAttributes;
// Logger for info/warning/errors
class Logger : public ILogger {
void log(Severity severity, const char* msg) NOEXCEPT override {
src/Yolo3Detection.cpp
+35 -16
View File
@@ -3,7 +3,7 @@
namespace tk { namespace dnn {
bool Yolo3Detection::init(const std::string& tensor_path, const int n_classes, const int n_batches, const float conf_thresh) {
bool Yolo3Detection::init(const std::string& tensor_path,const std::string& cfg_path,const std::string& name_path,const int n_classes, const int n_batches, const float conf_thresh) {
//convert network to tensorRT
std::cout<<(tensor_path).c_str()<<"\n";
@@ -14,28 +14,42 @@ bool Yolo3Detection::init(const std::string& tensor_path, const int n_classes, c
tk::dnn::dataDim_t idim = netRT->input_dim;
idim.n = nBatches;
std::vector<int> yolosLine = noYolosLine(cfg_path);
noYolos = yolosLine;
int channels,height,width;
loadYoloInitInfo(channels,width,height,cfg_path);
if(netRT->pluginFactory->n_yolos < 2 ) {
if(yolosLine.size() < 2 ) {
FatalError("this is not yolo3");
}
for(int i=0; i<netRT->pluginFactory->n_yolos; i++) {
YoloRT *yRT = netRT->pluginFactory->yolos[i];
classes = yRT->classes;
num = yRT->num;
nMasks = yRT->n_masks;
for(int i=0; i<noYolos.size(); i++) {
std::vector<float> maskTemp,anchorsTemp;
std::vector<std::string> classNamesTemp;
int classes,nms_kind,coords,numTemp;
float nmsthresh;
loadYoloInfo(cfg_path,yolosLine[i],maskTemp,anchorsTemp,numTemp,classes,nmsthresh,nms_kind,coords);
classNamesTemp = darknetReadNames(name_path);
num = numTemp/maskTemp.size();
nMasks = maskTemp.size();
dnnType* maskTempF;
dnnType* biasTempF;
maskTempF = maskTemp.data();
biasTempF = anchorsTemp.data();
// make a yolo layer to interpret predictions
yolo[i] = new tk::dnn::Yolo(nullptr, classes, nMasks, ""); // yolo without input and bias
yolo[i]->mask_h = new dnnType[nMasks];
yolo[i]->bias_h = new dnnType[num*nMasks*2];
memcpy(yolo[i]->mask_h, yRT->mask, sizeof(dnnType)*nMasks);
memcpy(yolo[i]->bias_h, yRT->bias, sizeof(dnnType)*num*nMasks*2);
yolo[i]->input_dim = yolo[i]->output_dim = tk::dnn::dataDim_t(1, yRT->c, yRT->h, yRT->w);
yolo[i]->classesNames = yRT->classesNames;
yolo[i]->nms_thresh = yRT->nms_thresh;
yolo[i]->nsm_kind = (tk::dnn::Yolo::nmsKind_t) yRT->nms_kind;
yolo[i]->new_coords = yRT->new_coords;
memcpy(yolo[i]->mask_h, maskTempF, sizeof(dnnType)*nMasks);
memcpy(yolo[i]->bias_h, biasTempF, sizeof(dnnType)*num*nMasks*2);
yolo[i]->input_dim = yolo[i]->output_dim = tk::dnn::dataDim_t(1, channels, height, width);
yolo[i]->classesNames = classNamesTemp;
yolo[i]->nms_thresh = nmsthresh;
yolo[i]->nsm_kind = (tk::dnn::Yolo::nmsKind_t) nms_kind;
yolo[i]->new_coords = coords;
}
dets = tk::dnn::Yolo::allocateDetections(tk::dnn::Yolo::MAX_DETECTIONS, classes);
@@ -94,10 +108,15 @@ void Yolo3Detection::preprocess(cv::Mat &frame, const int bi){
void Yolo3Detection::postprocess(const int bi, const bool mAP){
//get yolo outputs
if(noYolos.size() < 2){
FatalError("YOLOS WRONG!!");
}
std::vector<float *> rt_out;
//dnnType *rt_out[netRT->pluginFactory->n_yolos];
for(int i=0; i<netRT->pluginFactory->n_yolos; i++)
for(int i=0; i<noYolos.size(); i++)
rt_out.push_back((dnnType*)netRT->buffersRT[i+1] + netRT->buffersDIM[i+1].tot()*bi);
float x_ratio = float(originalSize[bi].width) / float(netRT->input_dim.w);
@@ -105,7 +124,7 @@ void Yolo3Detection::postprocess(const int bi, const bool mAP){
// compute dets
nDets = 0;
for(int i=0; i<netRT->pluginFactory->n_yolos; i++) {
for(int i=0; i<noYolos.size(); i++) {
yolo[i]->dstData = rt_out[i];
yolo[i]->computeDetections(dets, nDets, netRT->input_dim.w, netRT->input_dim.h, confThreshold, yolo[i]->new_coords);
}
tests/mnist/test_mnistRT.cpp
+1 -1
View File
@@ -128,7 +128,7 @@ int main() {
builder->setMaxBatchSize(1);
config->setMaxWorkspaceSize(1 << 20);
auto engine = builder->buildCudaEngine(*network);
auto engine = builder->buildEngineWithConfig(*network,*config);
// we don't need the network any more
network->destroy();