Skip to content

yolox_ros_cpp inference speed up. #24

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Ar-Ray-code merged 3 commits into from
Jun 14, 2022
Merged

yolox_ros_cpp inference speed up. #24

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c106a24
update yolox_cpp.
fateshelled Jun 13, 2022
9523998
update yolox_cpp
fateshelled Jun 13, 2022
06bc8d8
update yolox_ros_cpp
fateshelled Jun 14, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 1 addition & 6 deletions yolox_ros_cpp/yolox_cpp/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,11 +25,6 @@ set(ENABLE_TENSORRT OFF)
# find dependencies
find_package(ament_cmake REQUIRED)
find_package(OpenCV REQUIRED)
find_package(OpenMP)
if(OpenMP_FOUND)
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
endif()

if(YOLOX_USE_OPENVINO)
find_package(InferenceEngine)
Expand Down Expand Up @@ -139,7 +134,7 @@ install(
DIRECTORY include/
DESTINATION include
)

if(BUILD_TESTING)
find_package(ament_lint_auto REQUIRED)
# the following line skips the linter which checks for copyrights
Expand Down
73 changes: 34 additions & 39 deletions yolox_ros_cpp/yolox_cpp/include/yolox_cpp/core.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ namespace yolox_cpp{
:nms_thresh_(nms_th), bbox_conf_thresh_(conf_th),
model_version_(model_version)
{}
virtual std::vector<Object> inference(cv::Mat frame) = 0;
virtual std::vector<Object> inference(const cv::Mat& frame) = 0;
protected:
int input_w_;
int input_h_;
Expand All @@ -46,7 +46,8 @@ namespace yolox_cpp{
const std::vector<int> strides_ = {8, 16, 32};
std::vector<GridAndStride> grid_strides_;

cv::Mat static_resize(cv::Mat& img) {
cv::Mat static_resize(const cv::Mat& img)
{
float r = std::min(input_w_ / (img.cols*1.0), input_h_ / (img.rows*1.0));
// r = std::min(r, 1.0f);
int unpad_w = r * img.cols;
Expand All @@ -58,31 +59,31 @@ namespace yolox_cpp{
return out;
}

void blobFromImage(cv::Mat& img, float *blob_data){
void blobFromImage(const cv::Mat& img, float *blob_data)
{
int channels = 3;
int img_h = img.rows;
int img_w = img.cols;
if(this->model_version_=="0.1.0"){
for (size_t c = 0; c < channels; c++)
for (size_t c = 0; c < channels; ++c)
{
for (size_t h = 0; h < img_h; h++)
for (size_t h = 0; h < img_h; ++h)
{
for (size_t w = 0; w < img_w; w++)
for (size_t w = 0; w < img_w; ++w)
{
blob_data[c * img_w * img_h + h * img_w + w] =
((float)img.at<cv::Vec3b>(h, w)[c] / 255.0 - this->mean_[c]) / this->std_[c];
((float)img.ptr<cv::Vec3b>(h)[w][c]/ 255.0 - this->mean_[c]) / this->std_[c];
}
}
}
}else{
for (size_t c = 0; c < channels; c++)
for (size_t c = 0; c < channels; ++c)
{
for (size_t h = 0; h < img_h; h++)
for (size_t h = 0; h < img_h; ++h)
{
for (size_t w = 0; w < img_w; w++)
for (size_t w = 0; w < img_w; ++w)
{
blob_data[c * img_w * img_h + h * img_w + w] =
(float)img.at<cv::Vec3b>(h, w)[c]; // 0.1.1rc0 or later
blob_data[c * img_w * img_h + h * img_w + w] = (float)img.ptr<cv::Vec3b>(h)[w][c]; // 0.1.1rc0 or later
}
}
}
Expand All @@ -95,9 +96,9 @@ namespace yolox_cpp{
{
int num_grid_w = target_w / stride;
int num_grid_h = target_h / stride;
for (int g1 = 0; g1 < num_grid_h; g1++)
for (int g1 = 0; g1 < num_grid_h; ++g1)
{
for (int g0 = 0; g0 < num_grid_w; g0++)
for (int g0 = 0; g0 < num_grid_w; ++g0)
{
grid_strides.push_back((GridAndStride){g0, g1, stride});
}
Expand All @@ -109,7 +110,7 @@ namespace yolox_cpp{
{
const int num_anchors = grid_strides.size();

for (int anchor_idx = 0; anchor_idx < num_anchors; anchor_idx++)
for (int anchor_idx = 0; anchor_idx < num_anchors; ++anchor_idx)
{
const int grid0 = grid_strides[anchor_idx].grid0;
const int grid1 = grid_strides[anchor_idx].grid1;
Expand All @@ -120,7 +121,7 @@ namespace yolox_cpp{
float box_objectness = feat_ptr[basic_pos + 4];
int class_id = 0;
float max_class_score = 0.0;
for (int class_idx = 0; class_idx < num_classes_; class_idx++)
for (int class_idx = 0; class_idx < num_classes_; ++class_idx)
{
float box_cls_score = feat_ptr[basic_pos + 5 + class_idx];
float box_prob = box_objectness * box_cls_score;
Expand Down Expand Up @@ -168,32 +169,21 @@ namespace yolox_cpp{
while (i <= j)
{
while (faceobjects[i].prob > p)
i++;
++i;

while (faceobjects[j].prob < p)
j--;
--j;

if (i <= j)
{
// swap
std::swap(faceobjects[i], faceobjects[j]);

i++;
j--;
}
}

#pragma omp parallel sections
{
#pragma omp section
{
if (left < j) qsort_descent_inplace(faceobjects, left, j);
}
#pragma omp section
{
if (i < right) qsort_descent_inplace(faceobjects, i, right);
++i;
--j;
}
}
if (left < j) qsort_descent_inplace(faceobjects, left, j);
if (i < right) qsort_descent_inplace(faceobjects, i, right);
}

void qsort_descent_inplace(std::vector<Object>& objects)
Expand All @@ -211,17 +201,18 @@ namespace yolox_cpp{
const int n = faceobjects.size();

std::vector<float> areas(n);
for (int i = 0; i < n; i++)
for (int i = 0; i < n; ++i)
{
areas[i] = faceobjects[i].rect.area();
}

for (int i = 0; i < n; i++)
for (int i = 0; i < n; ++i)
{
const Object& a = faceobjects[i];
const int picked_size = picked.size();

int keep = 1;
for (int j = 0; j < (int)picked.size(); j++)
for (int j = 0; j < picked_size; ++j)
{
const Object& b = faceobjects[picked[j]];

Expand All @@ -238,19 +229,23 @@ namespace yolox_cpp{
}
}

void decode_outputs(const float* prob, const std::vector<GridAndStride> grid_strides, std::vector<Object>& objects, const float bbox_conf_thresh, const float scale, const int img_w, const int img_h) {
std::vector<Object> proposals;
void decode_outputs(const float* prob, const std::vector<GridAndStride> grid_strides,
std::vector<Object>& objects, const float bbox_conf_thresh,
const float scale, const int img_w, const int img_h)
{

std::vector<Object> proposals;
generate_yolox_proposals(grid_strides, prob, bbox_conf_thresh, proposals);

qsort_descent_inplace(proposals);

std::vector<int> picked;
nms_sorted_bboxes(proposals, picked, nms_thresh_);

int count = picked.size();
objects.resize(count);

for (int i = 0; i < count; i++)
for (int i = 0; i < count; ++i)
{
objects[i] = proposals[picked[i]];

Expand Down
2 changes: 1 addition & 1 deletion yolox_ros_cpp/yolox_cpp/include/yolox_cpp/yolox_openvino.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@ namespace yolox_cpp{
public:
YoloXOpenVINO(file_name_t path_to_model, std::string device_name,
float nms_th=0.45, float conf_th=0.3, std::string model_version="0.1.1rc0");
std::vector<Object> inference(cv::Mat frame) override;
std::vector<Object> inference(const cv::Mat& frame) override;

private:
std::string device_name_;
Expand Down
2 changes: 1 addition & 1 deletion yolox_ros_cpp/yolox_cpp/include/yolox_cpp/yolox_tensorrt.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@ namespace yolox_cpp{
public:
YoloXTensorRT(file_name_t path_to_engine, int device=0,
float nms_th=0.45, float conf_th=0.3, std::string model_version="0.1.1rc0");
std::vector<Object> inference(cv::Mat frame) override;
std::vector<Object> inference(const cv::Mat& frame) override;

private:
void doInference(float* input, float* output);
Expand Down
11 changes: 3 additions & 8 deletions yolox_ros_cpp/yolox_cpp/src/yolox_openvino.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,7 +62,8 @@ namespace yolox_cpp{
generate_grids_and_stride(this->input_w_, this->input_h_, this->strides_, this->grid_strides_);
}

std::vector<Object> YoloXOpenVINO::inference(cv::Mat frame){
std::vector<Object> YoloXOpenVINO::inference(const cv::Mat& frame)
{
// preprocess
cv::Mat pr_img = static_resize(frame);
InferenceEngine::Blob::Ptr imgBlob = infer_request_.GetBlob(input_name_);
Expand All @@ -79,12 +80,7 @@ namespace yolox_cpp{

// do inference
/* Running the request synchronously */
try{
infer_request_.Infer();
}catch (const std::exception& ex) {
std::cerr << ex.what() << std::endl;
return {};
}
infer_request_.Infer();

// Process output
const InferenceEngine::Blob::Ptr output_blob = infer_request_.GetBlob(output_name_);
Expand All @@ -100,7 +96,6 @@ namespace yolox_cpp{
const float* net_pred = moutputHolder.as<const PrecisionTrait<Precision::FP32>::value_type*>();

float scale = std::min(input_w_ / (frame.cols*1.0), input_h_ / (frame.rows*1.0));

std::vector<Object> objects;
decode_outputs(net_pred, this->grid_strides_, objects, this->bbox_conf_thresh_, scale, frame.cols, frame.rows);
return objects;
Expand Down
9 changes: 6 additions & 3 deletions yolox_ros_cpp/yolox_cpp/src/yolox_tensorrt.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,7 @@ namespace yolox_cpp{

auto out_dims = this->engine_->getBindingDimensions(1);
this->output_size_ = 1;
for(int j=0;j<out_dims.nbDims;j++) {
for(int j=0; j<out_dims.nbDims; ++j) {
this->output_size_ *= out_dims.d[j];
}

Expand All @@ -57,7 +57,9 @@ namespace yolox_cpp{
// Prepare GridAndStrides
generate_grids_and_stride(this->input_w_, this->input_h_, this->strides_, this->grid_strides_);
}
std::vector<Object> YoloXTensorRT::inference(cv::Mat frame){

std::vector<Object> YoloXTensorRT::inference(const cv::Mat& frame)
{
// preprocess
auto pr_img = static_resize(frame);
float* input_blob = new float[pr_img.total()*3];
Expand All @@ -77,7 +79,8 @@ namespace yolox_cpp{
return objects;
}

void YoloXTensorRT::doInference(float* input, float* output) {
void YoloXTensorRT::doInference(float* input, float* output)
{
// Pointers to input and output device buffers to pass to engine.
// Engine requires exactly IEngine::getNbBindings() number of buffers.
void* buffers[2];
Expand Down
27 changes: 16 additions & 11 deletions yolox_ros_cpp/yolox_ros_cpp/src/yolox_ros_cpp.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,8 +7,8 @@ namespace yolox_ros_cpp{
{}

YoloXNode::YoloXNode(const std::string &node_name, const rclcpp::NodeOptions& options)
: rclcpp::Node("yolox_ros_cpp", node_name, options){

: rclcpp::Node("yolox_ros_cpp", node_name, options)
{
RCLCPP_INFO(this->get_logger(), "initialize");
this->initializeParameter();

Expand Down Expand Up @@ -52,13 +52,14 @@ namespace yolox_ros_cpp{

}

void YoloXNode::initializeParameter(){
void YoloXNode::initializeParameter()
{
this->declare_parameter<bool>("imshow_isshow", true);
this->declare_parameter<std::string>("model_path", "/root/ros2_ws/src/YOLOX-ROS/weights/tensorrt/YOLOX_outputs/nano/model_trt.engine");
this->declare_parameter<std::string>("model_path", "src/YOLOX-ROS/weights/openvino/yolox_tiny.xml");
this->declare_parameter<float>("conf", 0.3f);
this->declare_parameter<float>("nms", 0.45f);
this->declare_parameter<std::string>("device", "0");
this->declare_parameter<std::string>("model_type", "tensorrt");
this->declare_parameter<std::string>("device", "CPU");
this->declare_parameter<std::string>("model_type", "openvino");
this->declare_parameter<std::string>("model_version", "0.1.1rc0");
this->declare_parameter<std::string>("src_image_topic_name", "image_raw");
this->declare_parameter<std::string>("publish_image_topic_name", "yolox/image_raw");
Expand Down Expand Up @@ -86,14 +87,20 @@ namespace yolox_ros_cpp{
RCLCPP_INFO(this->get_logger(), "Set parameter publish_image_topic_name: '%s'", this->publish_image_topic_name_.c_str());

}
void YoloXNode::colorImageCallback(const sensor_msgs::msg::Image::ConstSharedPtr& ptr){
void YoloXNode::colorImageCallback(const sensor_msgs::msg::Image::ConstSharedPtr& ptr)
{
auto img = cv_bridge::toCvCopy(ptr, "bgr8");
cv::Mat frame = img->image;

// fps
auto now = std::chrono::system_clock::now();

auto objects = this->yolox_->inference(frame);

auto end = std::chrono::system_clock::now();
auto elapsed = std::chrono::duration_cast<std::chrono::milliseconds>(end - now);
RCLCPP_INFO(this->get_logger(), "Inference: %f FPS", 1000.0f / elapsed.count());

yolox_cpp::utils::draw_objects(frame, objects);
if(this->imshow_){
cv::imshow(this->WINDOW_NAME_, frame);
Expand All @@ -110,11 +117,9 @@ namespace yolox_ros_cpp{
pub_img = cv_bridge::CvImage(img->header, "bgr8", frame).toImageMsg();
this->pub_image_.publish(pub_img);

auto end = std::chrono::system_clock::now();
auto elapsed = std::chrono::duration_cast<std::chrono::milliseconds>(end - now);
RCLCPP_INFO(this->get_logger(), "fps: %f", 1000.0f / elapsed.count());
}
bboxes_ex_msgs::msg::BoundingBoxes YoloXNode::objects_to_bboxes(cv::Mat frame, std::vector<yolox_cpp::Object> objects, std_msgs::msg::Header header){
bboxes_ex_msgs::msg::BoundingBoxes YoloXNode::objects_to_bboxes(cv::Mat frame, std::vector<yolox_cpp::Object> objects, std_msgs::msg::Header header)
{
bboxes_ex_msgs::msg::BoundingBoxes boxes;
boxes.header = header;
for(auto obj: objects){
Expand Down