xclele/yolov10
1
0
Fork 0
mirror of https://github.com/THU-MIG/yolov10.git synced 2025-05-23 13:34:23 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add C++ Classify inference example (#6868)

Browse Source 
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>
This commit is contained in:
DennisJ 2023-12-10 23:41:24 +08:00 committed by GitHub
parent 1b37a13131
commit b62b20d517
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 380 additions and 223 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
examples/YOLOv8-ONNXRuntime-CPP/README.md
View File

@ -13,6 +13,10 @@ This example demonstrates how to perform inference using YOLOv8 in C++ with ONNX
- Faster than OpenCV's DNN inference on both CPU and GPU. - Faster than OpenCV's DNN inference on both CPU and GPU.
- Supports FP32 and FP16 CUDA acceleration. - Supports FP32 and FP16 CUDA acceleration.
## Note :coffee:
1.~~This repository should also work for YOLOv5, which needs a permute operator for the output of the YOLOv5 model, but this has not been implemented yet.~~ Benefit for ultralytics's latest release,a `Transpose` op is added to the Yolov8 model,while make v8 and v5 has the same output shape.Therefore,you can inference your yolov5/v7/v8 via this project.
## Exporting YOLOv8 Models 📦 ## Exporting YOLOv8 Models 📦
To export YOLOv8 models, use the following Python script: To export YOLOv8 models, use the following Python script:
@ -33,6 +37,17 @@ Alternatively, you can use the following command for exporting the model in the
yolo export model=yolov8n.pt opset=12 simplify=True dynamic=False format=onnx imgsz=640,640 yolo export model=yolov8n.pt opset=12 simplify=True dynamic=False format=onnx imgsz=640,640
``` ```
## Exporting YOLOv8 FP16 Models 📦
```python
import onnx
from onnxconverter_common import float16
model = onnx.load(R'YOUR_ONNX_PATH')
model_fp16 = float16.convert_float_to_float16(model)
onnx.save(model_fp16, R'YOUR_FP16_ONNX_PATH')
```
## Download COCO.yaml file 📂 ## Download COCO.yaml file 📂
In order to run example, you also need to download coco.yaml. You can download the file manually from [here](https://raw.githubusercontent.com/ultralytics/ultralytics/main/ultralytics/cfg/datasets/coco.yaml) In order to run example, you also need to download coco.yaml. You can download the file manually from [here](https://raw.githubusercontent.com/ultralytics/ultralytics/main/ultralytics/cfg/datasets/coco.yaml)
@ -79,16 +94,15 @@ make
## Usage 🚀 ## Usage 🚀
```c++ ```c++
// CPU inference //change your param as you like
DCSP_INIT_PARAM params{ model_path, YOLO_ORIGIN_V8, {imgsz_w, imgsz_h}, 0.1, 0.5, false}; //Pay attention to your device and the onnx model type(fp32 or fp16)
// GPU inference DL_INIT_PARAM params;
DCSP_INIT_PARAM params{ model_path, YOLO_ORIGIN_V8, {imgsz_w, imgsz_h}, 0.1, 0.5, true}; params.rectConfidenceThreshold = 0.1;
// Load your image params.iouThreshold = 0.5;
cv::Mat img = cv::imread(img_path); params.modelPath = "yolov8n.onnx";
// Init Inference Session params.imgSize = { 640, 640 };
char* ret = yoloDetector->CreateSession(params); params.cudaEnable = true;
params.modelType = YOLO_DETECT_V8;
ret = yoloDetector->RunSession(img, res); yoloDetector->CreateSession(params);
Detector(yoloDetector);
``` ```
This repository should also work for YOLOv5, which needs a permute operator for the output of the YOLOv5 model, but this has not been implemented yet.

219
examples/YOLOv8-ONNXRuntime-CPP/inference.cpp
View File

@ -2,13 +2,13 @@
#include <regex> #include <regex>
#define benchmark #define benchmark
#define min(a,b) (((a) < (b)) ? (a) : (b))
DCSP_CORE::DCSP_CORE() { YOLO_V8::YOLO_V8() {
} }
DCSP_CORE::~DCSP_CORE() { YOLO_V8::~YOLO_V8() {
delete session; delete session;
} }
@ -22,14 +22,17 @@ namespace Ort
template<typename T> template<typename T>
char *BlobFromImage(cv::Mat &iImg, T &iBlob) { char* BlobFromImage(cv::Mat& iImg, T& iBlob) {
int channels = iImg.channels(); int channels = iImg.channels();
int imgHeight = iImg.rows; int imgHeight = iImg.rows;
int imgWidth = iImg.cols; int imgWidth = iImg.cols;
for (int c = 0; c < channels; c++) { for (int c = 0; c < channels; c++)
for (int h = 0; h < imgHeight; h++) { {
for (int w = 0; w < imgWidth; w++) { for (int h = 0; h < imgHeight; h++)
{
for (int w = 0; w < imgWidth; w++)
{
iBlob[c * imgWidth * imgHeight + h * imgWidth + w] = typename std::remove_pointer<T>::type( iBlob[c * imgWidth * imgHeight + h * imgWidth + w] = typename std::remove_pointer<T>::type(
(iImg.at<cv::Vec3b>(h, w)[c]) / 255.0f); (iImg.at<cv::Vec3b>(h, w)[c]) / 255.0f);
} }
@ -39,7 +42,7 @@ char *BlobFromImage(cv::Mat &iImg, T &iBlob) {
} }
char* DL_CORE::PreProcess(cv::Mat& iImg, std::vector<int> iImgSize, cv::Mat& oImg) char* YOLO_V8::PreProcess(cv::Mat& iImg, std::vector<int> iImgSize, cv::Mat& oImg)
{ {
if (iImg.channels() == 3) if (iImg.channels() == 3)
{ {
@ -51,6 +54,13 @@ char* DL_CORE::PreProcess(cv::Mat& iImg, std::vector<int> iImgSize, cv::Mat& oIm
cv::cvtColor(iImg, oImg, cv::COLOR_GRAY2RGB); cv::cvtColor(iImg, oImg, cv::COLOR_GRAY2RGB);
} }
switch (modelType)
{
case YOLO_DETECT_V8:
case YOLO_POSE:
case YOLO_DETECT_V8_HALF:
case YOLO_POSE_V8_HALF://LetterBox
{
if (iImg.cols >= iImg.rows) if (iImg.cols >= iImg.rows)
{ {
resizeScales = iImg.cols / (float)iImgSize.at(0); resizeScales = iImg.cols / (float)iImgSize.at(0);
@ -64,94 +74,116 @@ char* DL_CORE::PreProcess(cv::Mat& iImg, std::vector<int> iImgSize, cv::Mat& oIm
cv::Mat tempImg = cv::Mat::zeros(iImgSize.at(0), iImgSize.at(1), CV_8UC3); cv::Mat tempImg = cv::Mat::zeros(iImgSize.at(0), iImgSize.at(1), CV_8UC3);
oImg.copyTo(tempImg(cv::Rect(0, 0, oImg.cols, oImg.rows))); oImg.copyTo(tempImg(cv::Rect(0, 0, oImg.cols, oImg.rows)));
oImg = tempImg; oImg = tempImg;
break;
}
case YOLO_CLS://CenterCrop
{
int h = iImg.rows;
int w = iImg.cols;
int m = min(h, w);
int top = (h - m) / 2;
int left = (w - m) / 2;
cv::resize(oImg(cv::Rect(left, top, m, m)), oImg, cv::Size(iImgSize.at(0), iImgSize.at(1)));
break;
}
}
return RET_OK; return RET_OK;
} }
char *DCSP_CORE::CreateSession(DCSP_INIT_PARAM &iParams) { char* YOLO_V8::CreateSession(DL_INIT_PARAM& iParams) {
char *Ret = RET_OK; char* Ret = RET_OK;
std::regex pattern("[\u4e00-\u9fa5]"); std::regex pattern("[\u4e00-\u9fa5]");
bool result = std::regex_search(iParams.ModelPath, pattern); bool result = std::regex_search(iParams.modelPath, pattern);
if (result) { if (result)
Ret = "[DCSP_ONNX]:Model path error.Change your model path without chinese characters."; {
Ret = "[YOLO_V8]:Your model path is error.Change your model path without chinese characters.";
std::cout << Ret << std::endl; std::cout << Ret << std::endl;
return Ret; return Ret;
} }
try { try
rectConfidenceThreshold = iParams.RectConfidenceThreshold; {
rectConfidenceThreshold = iParams.rectConfidenceThreshold;
iouThreshold = iParams.iouThreshold; iouThreshold = iParams.iouThreshold;
imgSize = iParams.imgSize; imgSize = iParams.imgSize;
modelType = iParams.ModelType; modelType = iParams.modelType;
env = Ort::Env(ORT_LOGGING_LEVEL_WARNING, "Yolo"); env = Ort::Env(ORT_LOGGING_LEVEL_WARNING, "Yolo");
Ort::SessionOptions sessionOption; Ort::SessionOptions sessionOption;
if (iParams.CudaEnable) { if (iParams.cudaEnable)
cudaEnable = iParams.CudaEnable; {
cudaEnable = iParams.cudaEnable;
OrtCUDAProviderOptions cudaOption; OrtCUDAProviderOptions cudaOption;
cudaOption.device_id = 0; cudaOption.device_id = 0;
sessionOption.AppendExecutionProvider_CUDA(cudaOption); sessionOption.AppendExecutionProvider_CUDA(cudaOption);
} }
sessionOption.SetGraphOptimizationLevel(GraphOptimizationLevel::ORT_ENABLE_ALL); sessionOption.SetGraphOptimizationLevel(GraphOptimizationLevel::ORT_ENABLE_ALL);
sessionOption.SetIntraOpNumThreads(iParams.IntraOpNumThreads); sessionOption.SetIntraOpNumThreads(iParams.intraOpNumThreads);
sessionOption.SetLogSeverityLevel(iParams.LogSeverityLevel); sessionOption.SetLogSeverityLevel(iParams.logSeverityLevel);
#ifdef _WIN32 #ifdef _WIN32
int ModelPathSize = MultiByteToWideChar(CP_UTF8, 0, iParams.ModelPath.c_str(), static_cast<int>(iParams.ModelPath.length()), nullptr, 0); int ModelPathSize = MultiByteToWideChar(CP_UTF8, 0, iParams.modelPath.c_str(), static_cast<int>(iParams.modelPath.length()), nullptr, 0);
wchar_t* wide_cstr = new wchar_t[ModelPathSize + 1]; wchar_t* wide_cstr = new wchar_t[ModelPathSize + 1];
MultiByteToWideChar(CP_UTF8, 0, iParams.ModelPath.c_str(), static_cast<int>(iParams.ModelPath.length()), wide_cstr, ModelPathSize); MultiByteToWideChar(CP_UTF8, 0, iParams.modelPath.c_str(), static_cast<int>(iParams.modelPath.length()), wide_cstr, ModelPathSize);
wide_cstr[ModelPathSize] = L'\0'; wide_cstr[ModelPathSize] = L'\0';
const wchar_t* modelPath = wide_cstr; const wchar_t* modelPath = wide_cstr;
#else #else
const char *modelPath = iParams.ModelPath.c_str(); const char* modelPath = iParams.ModelPath.c_str();
#endif // _WIN32 #endif // _WIN32
session = new Ort::Session(env, modelPath, sessionOption); session = new Ort::Session(env, modelPath, sessionOption);
Ort::AllocatorWithDefaultOptions allocator; Ort::AllocatorWithDefaultOptions allocator;
size_t inputNodesNum = session->GetInputCount(); size_t inputNodesNum = session->GetInputCount();
for (size_t i = 0; i < inputNodesNum; i++) { for (size_t i = 0; i < inputNodesNum; i++)
{
Ort::AllocatedStringPtr input_node_name = session->GetInputNameAllocated(i, allocator); Ort::AllocatedStringPtr input_node_name = session->GetInputNameAllocated(i, allocator);
char *temp_buf = new char[50]; char* temp_buf = new char[50];
strcpy(temp_buf, input_node_name.get()); strcpy(temp_buf, input_node_name.get());
inputNodeNames.push_back(temp_buf); inputNodeNames.push_back(temp_buf);
} }
size_t OutputNodesNum = session->GetOutputCount(); size_t OutputNodesNum = session->GetOutputCount();
for (size_t i = 0; i < OutputNodesNum; i++) { for (size_t i = 0; i < OutputNodesNum; i++)
{
Ort::AllocatedStringPtr output_node_name = session->GetOutputNameAllocated(i, allocator); Ort::AllocatedStringPtr output_node_name = session->GetOutputNameAllocated(i, allocator);
char *temp_buf = new char[10]; char* temp_buf = new char[10];
strcpy(temp_buf, output_node_name.get()); strcpy(temp_buf, output_node_name.get());
outputNodeNames.push_back(temp_buf); outputNodeNames.push_back(temp_buf);
} }
options = Ort::RunOptions{nullptr}; options = Ort::RunOptions{ nullptr };
WarmUpSession(); WarmUpSession();
return RET_OK; return RET_OK;
} }
catch (const std::exception &e) { catch (const std::exception& e)
const char *str1 = "[DCSP_ONNX]:"; {
const char *str2 = e.what(); const char* str1 = "[YOLO_V8]:";
const char* str2 = e.what();
std::string result = std::string(str1) + std::string(str2); std::string result = std::string(str1) + std::string(str2);
char *merged = new char[result.length() + 1]; char* merged = new char[result.length() + 1];
std::strcpy(merged, result.c_str()); std::strcpy(merged, result.c_str());
std::cout << merged << std::endl; std::cout << merged << std::endl;
delete[] merged; delete[] merged;
return "[DCSP_ONNX]:Create session failed."; return "[YOLO_V8]:Create session failed.";
} }
} }
char *DCSP_CORE::RunSession(cv::Mat &iImg, std::vector<DCSP_RESULT> &oResult) { char* YOLO_V8::RunSession(cv::Mat& iImg, std::vector<DL_RESULT>& oResult) {
#ifdef benchmark #ifdef benchmark
clock_t starttime_1 = clock(); clock_t starttime_1 = clock();
#endif // benchmark #endif // benchmark
char *Ret = RET_OK; char* Ret = RET_OK;
cv::Mat processedImg; cv::Mat processedImg;
PreProcess(iImg, imgSize, processedImg); PreProcess(iImg, imgSize, processedImg);
if (modelType < 4) { if (modelType < 4)
float *blob = new float[processedImg.total() * 3]; {
float* blob = new float[processedImg.total() * 3];
BlobFromImage(processedImg, blob); BlobFromImage(processedImg, blob);
std::vector<int64_t> inputNodeDims = {1, 3, imgSize.at(0), imgSize.at(1)}; std::vector<int64_t> inputNodeDims = { 1, 3, imgSize.at(0), imgSize.at(1) };
TensorProcess(starttime_1, iImg, blob, inputNodeDims, oResult); TensorProcess(starttime_1, iImg, blob, inputNodeDims, oResult);
} else { }
else
{
#ifdef USE_CUDA #ifdef USE_CUDA
half* blob = new half[processedImg.total() * 3]; half* blob = new half[processedImg.total() * 3];
BlobFromImage(processedImg, blob); BlobFromImage(processedImg, blob);
@ -165,8 +197,8 @@ char *DCSP_CORE::RunSession(cv::Mat &iImg, std::vector<DCSP_RESULT> &oResult) {
template<typename N> template<typename N>
char *DCSP_CORE::TensorProcess(clock_t &starttime_1, cv::Mat &iImg, N &blob, std::vector<int64_t> &inputNodeDims, char* YOLO_V8::TensorProcess(clock_t& starttime_1, cv::Mat& iImg, N& blob, std::vector<int64_t>& inputNodeDims,
std::vector<DCSP_RESULT> &oResult) { std::vector<DL_RESULT>& oResult) {
Ort::Value inputTensor = Ort::Value::CreateTensor<typename std::remove_pointer<N>::type>( Ort::Value inputTensor = Ort::Value::CreateTensor<typename std::remove_pointer<N>::type>(
Ort::MemoryInfo::CreateCpu(OrtDeviceAllocator, OrtMemTypeCPU), blob, 3 * imgSize.at(0) * imgSize.at(1), Ort::MemoryInfo::CreateCpu(OrtDeviceAllocator, OrtMemTypeCPU), blob, 3 * imgSize.at(0) * imgSize.at(1),
inputNodeDims.data(), inputNodeDims.size()); inputNodeDims.data(), inputNodeDims.size());
@ -184,38 +216,46 @@ char *DCSP_CORE::TensorProcess(clock_t &starttime_1, cv::Mat &iImg, N &blob, std
std::vector<int64_t> outputNodeDims = tensor_info.GetShape(); std::vector<int64_t> outputNodeDims = tensor_info.GetShape();
auto output = outputTensor.front().GetTensorMutableData<typename std::remove_pointer<N>::type>(); auto output = outputTensor.front().GetTensorMutableData<typename std::remove_pointer<N>::type>();
delete blob; delete blob;
switch (modelType) { switch (modelType)
case 1://V8_ORIGIN_FP32
case 4://V8_ORIGIN_FP16
{ {
int strideNum = outputNodeDims[2]; case YOLO_DETECT_V8:
int signalResultNum = outputNodeDims[1]; case YOLO_DETECT_V8_HALF:
{
int strideNum = outputNodeDims[1];//8400
int signalResultNum = outputNodeDims[2];//84
std::vector<int> class_ids; std::vector<int> class_ids;
std::vector<float> confidences; std::vector<float> confidences;
std::vector<cv::Rect> boxes; std::vector<cv::Rect> boxes;
cv::Mat rawData; cv::Mat rawData;
if (modelType == 1) { if (modelType == YOLO_DETECT_V8)
{
// FP32 // FP32
rawData = cv::Mat(signalResultNum, strideNum, CV_32F, output); rawData = cv::Mat(strideNum, signalResultNum, CV_32F, output);
} else { }
else
{
// FP16 // FP16
rawData = cv::Mat(signalResultNum, strideNum, CV_16F, output); rawData = cv::Mat(strideNum, signalResultNum, CV_16F, output);
rawData.convertTo(rawData, CV_32F); rawData.convertTo(rawData, CV_32F);
} }
rawData = rawData.t(); //Note:
float *data = (float *) rawData.data; //ultralytics add transpose operator to the output of yolov8 model.which make yolov8/v5/v7 has same shape
//https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8n.pt
//rowData = rowData.t();
for (int i = 0; i < strideNum; ++i) { float* data = (float*)rawData.data;
float *classesScores = data + 4;
for (int i = 0; i < strideNum; ++i)
{
float* classesScores = data + 4;
cv::Mat scores(1, this->classes.size(), CV_32FC1, classesScores); cv::Mat scores(1, this->classes.size(), CV_32FC1, classesScores);
cv::Point class_id; cv::Point class_id;
double maxClassScore; double maxClassScore;
cv::minMaxLoc(scores, 0, &maxClassScore, 0, &class_id); cv::minMaxLoc(scores, 0, &maxClassScore, 0, &class_id);
if (maxClassScore > rectConfidenceThreshold) { if (maxClassScore > rectConfidenceThreshold)
{
confidences.push_back(maxClassScore); confidences.push_back(maxClassScore);
class_ids.push_back(class_id.x); class_ids.push_back(class_id.x);
float x = data[0]; float x = data[0];
float y = data[1]; float y = data[1];
float w = data[2]; float w = data[2];
@ -227,54 +267,68 @@ char *DCSP_CORE::TensorProcess(clock_t &starttime_1, cv::Mat &iImg, N &blob, std
int width = int(w * resizeScales); int width = int(w * resizeScales);
int height = int(h * resizeScales); int height = int(h * resizeScales);
boxes.emplace_back(left, top, width, height); boxes.push_back(cv::Rect(left, top, width, height));
} }
data += signalResultNum; data += signalResultNum;
} }
std::vector<int> nmsResult; std::vector<int> nmsResult;
cv::dnn::NMSBoxes(boxes, confidences, rectConfidenceThreshold, iouThreshold, nmsResult); cv::dnn::NMSBoxes(boxes, confidences, rectConfidenceThreshold, iouThreshold, nmsResult);
for (int i = 0; i < nmsResult.size(); ++i)
for (int i = 0; i < nmsResult.size(); ++i) { {
int idx = nmsResult[i]; int idx = nmsResult[i];
DCSP_RESULT result; DL_RESULT result;
result.classId = class_ids[idx]; result.classId = class_ids[idx];
result.confidence = confidences[idx]; result.confidence = confidences[idx];
result.box = boxes[idx]; result.box = boxes[idx];
oResult.push_back(result); oResult.push_back(result);
} }
#ifdef benchmark #ifdef benchmark
clock_t starttime_4 = clock(); clock_t starttime_4 = clock();
double pre_process_time = (double) (starttime_2 - starttime_1) / CLOCKS_PER_SEC * 1000; double pre_process_time = (double)(starttime_2 - starttime_1) / CLOCKS_PER_SEC * 1000;
double process_time = (double) (starttime_3 - starttime_2) / CLOCKS_PER_SEC * 1000; double process_time = (double)(starttime_3 - starttime_2) / CLOCKS_PER_SEC * 1000;
double post_process_time = (double) (starttime_4 - starttime_3) / CLOCKS_PER_SEC * 1000; double post_process_time = (double)(starttime_4 - starttime_3) / CLOCKS_PER_SEC * 1000;
if (cudaEnable) { if (cudaEnable)
std::cout << "[DCSP_ONNX(CUDA)]: " << pre_process_time << "ms pre-process, " << process_time {
<< "ms inference, " << post_process_time << "ms post-process." << std::endl; std::cout << "[YOLO_V8(CUDA)]: " << pre_process_time << "ms pre-process, " << process_time << "ms inference, " << post_process_time << "ms post-process." << std::endl;
} else { }
std::cout << "[DCSP_ONNX(CPU)]: " << pre_process_time << "ms pre-process, " << process_time else
<< "ms inference, " << post_process_time << "ms post-process." << std::endl; {
std::cout << "[YOLO_V8(CPU)]: " << pre_process_time << "ms pre-process, " << process_time << "ms inference, " << post_process_time << "ms post-process." << std::endl;
} }
#endif // benchmark #endif // benchmark
break; break;
} }
case YOLO_CLS:
{
DL_RESULT result;
for (int i = 0; i < this->classes.size(); i++)
{
result.classId = i;
result.confidence = output[i];
oResult.push_back(result);
}
break;
}
default:
std::cout << "[YOLO_V8]: " << "Not support model type." << std::endl;
} }
return RET_OK; return RET_OK;
} }
char *DCSP_CORE::WarmUpSession() { char* YOLO_V8::WarmUpSession() {
clock_t starttime_1 = clock(); clock_t starttime_1 = clock();
cv::Mat iImg = cv::Mat(cv::Size(imgSize.at(0), imgSize.at(1)), CV_8UC3); cv::Mat iImg = cv::Mat(cv::Size(imgSize.at(0), imgSize.at(1)), CV_8UC3);
cv::Mat processedImg; cv::Mat processedImg;
PreProcess(iImg, imgSize, processedImg); PreProcess(iImg, imgSize, processedImg);
if (modelType < 4) { if (modelType < 4)
float *blob = new float[iImg.total() * 3]; {
float* blob = new float[iImg.total() * 3];
BlobFromImage(processedImg, blob); BlobFromImage(processedImg, blob);
std::vector<int64_t> YOLO_input_node_dims = {1, 3, imgSize.at(0), imgSize.at(1)}; std::vector<int64_t> YOLO_input_node_dims = { 1, 3, imgSize.at(0), imgSize.at(1) };
Ort::Value input_tensor = Ort::Value::CreateTensor<float>( Ort::Value input_tensor = Ort::Value::CreateTensor<float>(
Ort::MemoryInfo::CreateCpu(OrtDeviceAllocator, OrtMemTypeCPU), blob, 3 * imgSize.at(0) * imgSize.at(1), Ort::MemoryInfo::CreateCpu(OrtDeviceAllocator, OrtMemTypeCPU), blob, 3 * imgSize.at(0) * imgSize.at(1),
YOLO_input_node_dims.data(), YOLO_input_node_dims.size()); YOLO_input_node_dims.data(), YOLO_input_node_dims.size());
@ -282,11 +336,14 @@ char *DCSP_CORE::WarmUpSession() {
outputNodeNames.size()); outputNodeNames.size());
delete[] blob; delete[] blob;
clock_t starttime_4 = clock(); clock_t starttime_4 = clock();
double post_process_time = (double) (starttime_4 - starttime_1) / CLOCKS_PER_SEC * 1000; double post_process_time = (double)(starttime_4 - starttime_1) / CLOCKS_PER_SEC * 1000;
if (cudaEnable) { if (cudaEnable)
std::cout << "[DCSP_ONNX(CUDA)]: " << "Cuda warm-up cost " << post_process_time << " ms. " << std::endl; {
std::cout << "[YOLO_V8(CUDA)]: " << "Cuda warm-up cost " << post_process_time << " ms. " << std::endl;
} }
} else { }
else
{
#ifdef USE_CUDA #ifdef USE_CUDA
half* blob = new half[iImg.total() * 3]; half* blob = new half[iImg.total() * 3];
BlobFromImage(processedImg, blob); BlobFromImage(processedImg, blob);
@ -298,7 +355,7 @@ char *DCSP_CORE::WarmUpSession() {
double post_process_time = (double)(starttime_4 - starttime_1) / CLOCKS_PER_SEC * 1000; double post_process_time = (double)(starttime_4 - starttime_1) / CLOCKS_PER_SEC * 1000;
if (cudaEnable) if (cudaEnable)
{ {
std::cout << "[DCSP_ONNX(CUDA)]: " << "Cuda warm-up cost " << post_process_time << " ms. " << std::endl; std::cout << "[YOLO_V8(CUDA)]: " << "Cuda warm-up cost " << post_process_time << " ms. " << std::endl;
} }
#endif #endif
} }

64
examples/YOLOv8-ONNXRuntime-CPP/inference.h
View File

@ -19,53 +19,59 @@
#endif #endif
enum MODEL_TYPE { enum MODEL_TYPE
{
//FLOAT32 MODEL //FLOAT32 MODEL
YOLO_ORIGIN_V5 = 0, YOLO_DETECT_V8 = 1,
YOLO_ORIGIN_V8 = 1,//only support v8 detector currently YOLO_POSE = 2,
YOLO_POSE_V8 = 2, YOLO_CLS = 3,
YOLO_CLS_V8 = 3,
YOLO_ORIGIN_V8_HALF = 4, //FLOAT16 MODEL
YOLO_DETECT_V8_HALF = 4,
YOLO_POSE_V8_HALF = 5, YOLO_POSE_V8_HALF = 5,
YOLO_CLS_V8_HALF = 6
}; };
typedef struct _DCSP_INIT_PARAM { typedef struct _DL_INIT_PARAM
std::string ModelPath; {
MODEL_TYPE ModelType = YOLO_ORIGIN_V8; std::string modelPath;
std::vector<int> imgSize = {640, 640}; MODEL_TYPE modelType = YOLO_DETECT_V8;
float RectConfidenceThreshold = 0.6; std::vector<int> imgSize = { 640, 640 };
float rectConfidenceThreshold = 0.6;
float iouThreshold = 0.5; float iouThreshold = 0.5;
bool CudaEnable = false; int keyPointsNum = 2;//Note:kpt number for pose
int LogSeverityLevel = 3; bool cudaEnable = false;
int IntraOpNumThreads = 1; int logSeverityLevel = 3;
} DCSP_INIT_PARAM; int intraOpNumThreads = 1;
} DL_INIT_PARAM;
typedef struct _DCSP_RESULT { typedef struct _DL_RESULT
{
int classId; int classId;
float confidence; float confidence;
cv::Rect box; cv::Rect box;
} DCSP_RESULT; std::vector<cv::Point2f> keyPoints;
} DL_RESULT;
class DCSP_CORE { class YOLO_V8
{
public: public:
DCSP_CORE(); YOLO_V8();
~DCSP_CORE(); ~YOLO_V8();
public: public:
char *CreateSession(DCSP_INIT_PARAM &iParams); char* CreateSession(DL_INIT_PARAM& iParams);
char *RunSession(cv::Mat &iImg, std::vector<DCSP_RESULT> &oResult); char* RunSession(cv::Mat& iImg, std::vector<DL_RESULT>& oResult);
char *WarmUpSession(); char* WarmUpSession();
template<typename N> template<typename N>
char *TensorProcess(clock_t &starttime_1, cv::Mat &iImg, N &blob, std::vector<int64_t> &inputNodeDims, char* TensorProcess(clock_t& starttime_1, cv::Mat& iImg, N& blob, std::vector<int64_t>& inputNodeDims,
std::vector<DCSP_RESULT> &oResult); std::vector<DL_RESULT>& oResult);
char* PreProcess(cv::Mat& iImg, std::vector<int> iImgSize, cv::Mat& oImg); char* PreProcess(cv::Mat& iImg, std::vector<int> iImgSize, cv::Mat& oImg);
@ -73,11 +79,11 @@ public:
private: private:
Ort::Env env; Ort::Env env;
Ort::Session *session; Ort::Session* session;
bool cudaEnable; bool cudaEnable;
Ort::RunOptions options; Ort::RunOptions options;
std::vector<const char *> inputNodeNames; std::vector<const char*> inputNodeNames;
std::vector<const char *> outputNodeNames; std::vector<const char*> outputNodeNames;
MODEL_TYPE modelType; MODEL_TYPE modelType;
std::vector<int> imgSize; std::vector<int> imgSize;

120
examples/YOLOv8-ONNXRuntime-CPP/main.cpp
View File

@ -3,18 +3,22 @@
#include "inference.h" #include "inference.h"
#include <filesystem> #include <filesystem>
#include <fstream> #include <fstream>
#include <random>
void file_iterator(DCSP_CORE *&p) { void Detector(YOLO_V8*& p) {
std::filesystem::path current_path = std::filesystem::current_path(); std::filesystem::path current_path = std::filesystem::current_path();
std::filesystem::path imgs_path = current_path / "images"; std::filesystem::path imgs_path = current_path / "images";
for (auto &i: std::filesystem::directory_iterator(imgs_path)) { for (auto& i : std::filesystem::directory_iterator(imgs_path))
if (i.path().extension() == ".jpg" || i.path().extension() == ".png" || i.path().extension() == ".jpeg") { {
if (i.path().extension() == ".jpg" || i.path().extension() == ".png" || i.path().extension() == ".jpeg")
{
std::string img_path = i.path().string(); std::string img_path = i.path().string();
cv::Mat img = cv::imread(img_path); cv::Mat img = cv::imread(img_path);
std::vector<DCSP_RESULT> res; std::vector<DL_RESULT> res;
p->RunSession(img, res); p->RunSession(img, res);
for (auto &re: res) { for (auto& re : res)
{
cv::RNG rng(cv::getTickCount()); cv::RNG rng(cv::getTickCount());
cv::Scalar color(rng.uniform(0, 256), rng.uniform(0, 256), rng.uniform(0, 256)); cv::Scalar color(rng.uniform(0, 256), rng.uniform(0, 256), rng.uniform(0, 256));
@ -53,10 +57,51 @@ void file_iterator(DCSP_CORE *&p) {
} }
} }
int read_coco_yaml(DCSP_CORE *&p) {
void Classifier(YOLO_V8*& p)
{
std::filesystem::path current_path = std::filesystem::current_path();
std::filesystem::path imgs_path = current_path;// / "images"
std::random_device rd;
std::mt19937 gen(rd());
std::uniform_int_distribution<int> dis(0, 255);
for (auto& i : std::filesystem::directory_iterator(imgs_path))
{
if (i.path().extension() == ".jpg" || i.path().extension() == ".png")
{
std::string img_path = i.path().string();
//std::cout << img_path << std::endl;
cv::Mat img = cv::imread(img_path);
std::vector<DL_RESULT> res;
char* ret = p->RunSession(img, res);
float positionY = 50;
for (int i = 0; i < res.size(); i++)
{
int r = dis(gen);
int g = dis(gen);
int b = dis(gen);
cv::putText(img, std::to_string(i) + ":", cv::Point(10, positionY), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(b, g, r), 2);
cv::putText(img, std::to_string(res.at(i).confidence), cv::Point(70, positionY), cv::FONT_HERSHEY_SIMPLEX, 1, cv::Scalar(b, g, r), 2);
positionY += 50;
}
cv::imshow("TEST_CLS", img);
cv::waitKey(0);
cv::destroyAllWindows();
//cv::imwrite("E:\\output\\" + std::to_string(k) + ".png", img);
}
}
}
int ReadCocoYaml(YOLO_V8*& p) {
// Open the YAML file // Open the YAML file
std::ifstream file("coco.yaml"); std::ifstream file("coco.yaml");
if (!file.is_open()) { if (!file.is_open())
{
std::cerr << "Failed to open file" << std::endl; std::cerr << "Failed to open file" << std::endl;
return 1; return 1;
} }
@ -64,17 +109,22 @@ int read_coco_yaml(DCSP_CORE *&p) {
// Read the file line by line // Read the file line by line
std::string line; std::string line;
std::vector<std::string> lines; std::vector<std::string> lines;
while (std::getline(file, line)) { while (std::getline(file, line))
{
lines.push_back(line); lines.push_back(line);
} }
// Find the start and end of the names section // Find the start and end of the names section
std::size_t start = 0; std::size_t start = 0;
std::size_t end = 0; std::size_t end = 0;
for (std::size_t i = 0; i < lines.size(); i++) { for (std::size_t i = 0; i < lines.size(); i++)
if (lines[i].find("names:") != std::string::npos) { {
if (lines[i].find("names:") != std::string::npos)
{
start = i + 1; start = i + 1;
} else if (start > 0 && lines[i].find(':') == std::string::npos) { }
else if (start > 0 && lines[i].find(':') == std::string::npos)
{
end = i; end = i;
break; break;
} }
@ -82,7 +132,8 @@ int read_coco_yaml(DCSP_CORE *&p) {
// Extract the names // Extract the names
std::vector<std::string> names; std::vector<std::string> names;
for (std::size_t i = start; i < end; i++) { for (std::size_t i = start; i < end; i++)
{
std::stringstream ss(lines[i]); std::stringstream ss(lines[i]);
std::string name; std::string name;
std::getline(ss, name, ':'); // Extract the number before the delimiter std::getline(ss, name, ':'); // Extract the number before the delimiter
@ -95,19 +146,48 @@ int read_coco_yaml(DCSP_CORE *&p) {
} }
int main() { void DetectTest()
DCSP_CORE *yoloDetector = new DCSP_CORE; {
std::string model_path = "yolov8n.onnx"; YOLO_V8* yoloDetector = new YOLO_V8;
read_coco_yaml(yoloDetector); ReadCocoYaml(yoloDetector);
DL_INIT_PARAM params;
params.rectConfidenceThreshold = 0.1;
params.iouThreshold = 0.5;
params.modelPath = "yolov8n.onnx";
params.imgSize = { 640, 640 };
#ifdef USE_CUDA #ifdef USE_CUDA
params.cudaEnable = true;
// GPU FP32 inference // GPU FP32 inference
DCSP_INIT_PARAM params{ model_path, YOLO_ORIGIN_V8, {640, 640}, 0.1, 0.5, true }; params.modelType = YOLO_DETECT_V8;
// GPU FP16 inference // GPU FP16 inference
// DCSP_INIT_PARAM params{ model_path, YOLO_ORIGIN_V8_HALF, {640, 640}, 0.1, 0.5, true }; //Note: change fp16 onnx model
//params.modelType = YOLO_DETECT_V8_HALF;
#else #else
// CPU inference // CPU inference
DCSP_INIT_PARAM params{model_path, YOLO_ORIGIN_V8, {640, 640}, 0.1, 0.5, false}; params.modelType = YOLO_DETECT_V8;
params.cudaEnable = false;
#endif #endif
yoloDetector->CreateSession(params); yoloDetector->CreateSession(params);
file_iterator(yoloDetector); Detector(yoloDetector);
}
void ClsTest()
{
YOLO_V8* yoloDetector = new YOLO_V8;
std::string model_path = "cls.onnx";
ReadCocoYaml(yoloDetector);
DL_INIT_PARAM params{ model_path, YOLO_CLS, {224, 224} };
yoloDetector->CreateSession(params);
Classifier(yoloDetector);
}
int main()
{
//DetectTest();
ClsTest();
} }