├── configs ├── yolov5-2.0 │ ├── .gitkeep │ ├── yolov5l.yaml │ ├── yolov5m.yaml │ ├── yolov5s.yaml │ ├── yolov5x.yaml │ ├── yolov5m.cfg │ ├── yolov5l.cfg │ ├── yolov5s.cfg │ └── yolov5x.cfg ├── yolov5-3.0 │ ├── .gitkeep │ ├── yolov5l.yaml │ ├── yolov5m.yaml │ ├── yolov5s.yaml │ ├── yolov5x.yaml │ ├── yolov5l.cfg │ ├── yolov5m.cfg │ ├── yolov5s.cfg │ └── yolov5x.cfg ├── dog.jpg ├── person.jpg ├── result.jpg ├── qq_group.png ├── qrcode.jpeg ├── yolo-trt.png ├── yolov5-4.0 │ ├── yolov5l.yaml │ ├── yolov5m.yaml │ ├── yolov5s.yaml │ ├── yolov5x.yaml │ ├── yolov5l.cfg │ ├── yolov5m.cfg │ ├── yolov5s.cfg │ └── yolov5x.cfg ├── yolov5-5.0 │ ├── yolov5l.yaml │ ├── yolov5m.yaml │ ├── yolov5s.yaml │ ├── yolov5x.yaml │ ├── yolov5l.cfg │ ├── yolov5m.cfg │ ├── yolov5s.cfg │ ├── yolov5x.cfg │ ├── yolov5l6.yaml │ ├── yolov5m6.yaml │ ├── yolov5s6.yaml │ ├── yolov5x6.yaml │ ├── yolov5l6.cfg │ ├── yolov5s6.cfg │ ├── yolov5m6.cfg │ └── yolov5x6.cfg ├── yolov5-6.0 │ ├── yolov5l.yaml │ ├── yolov5m.yaml │ ├── yolov5n.yaml │ ├── yolov5s.yaml │ ├── yolov5x.yaml │ ├── yolov5l.cfg │ ├── yolov5s.cfg │ ├── yolov5m.cfg │ ├── yolov5n.cfg │ └── yolov5x.cfg ├── yolov3-tiny.cfg └── yolov4-tiny.cfg ├── modules ├── decodeTensorCUDA.h ├── yolov4.h ├── class_detector.cpp ├── yolov5.h ├── class_detector.h ├── yolov5.cpp ├── yolov3.h ├── yolov2.h ├── yolo_config_parser.h ├── yolov4.cpp ├── ds_image.h ├── calibrator.h ├── kernel.cu ├── yoloplugin_lib.h ├── chunk.h ├── yolov3.cpp ├── yolov2.cpp ├── decodeTensorCUDA.cu ├── hardswish.h ├── mish.h ├── detect.h ├── class_yolo_detector.hpp ├── calibrator.cpp └── hardswish.cu ├── extra ├── API.h └── class_timer.hpp ├── .gitignore ├── sln ├── test_dll │ ├── test_dll.vcxproj.user │ ├── test_dll.vcxproj.filters │ └── test_dll.vcxproj ├── dll_detector │ ├── dll_detector.vcxproj.user │ ├── TensorRT-8.0.1.6_cuda-10.2.cudnn8.2.props │ └── dll_detector.vcxproj.filters ├── props │ ├── opencv412_release_x64.props │ ├── TensorRT-8.0.1.6_cuda-11.3_cudnn8.2.props │ ├── cuda100_x64_release.props │ ├── cuda11_x64_release.props │ ├── gflags_x64_release.props │ ├── TensorRT-5.1.5.0_cuda-10.0_cudnn-7.5.props │ ├── opencv330_release_x64.props │ ├── TensorRT-7.1.3.4_cuda-11.0_cudnn8.0.props │ ├── TensorRT-7.2.1.6_cuda-11.0.cudnn8.0.props │ ├── TensorRT-7.2.3.4_cuda-11.1.cudnn8.1.props │ ├── dll_debug_release_x64.props │ ├── console_debug_release_x64 .props │ └── opencv412_release_x64_vs2017.props └── sln.sln ├── scripts ├── CMakeLists.txt ├── download.py └── yaml2cfg.py ├── LICENSE ├── CMakeLists.txt ├── samples └── sample_detector.cpp └── README.md /configs/yolov5-2.0/.gitkeep: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /configs/yolov5-3.0/.gitkeep: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /configs/dog.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/enazoe/yolo-tensorrt/HEAD/configs/dog.jpg -------------------------------------------------------------------------------- /configs/person.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/enazoe/yolo-tensorrt/HEAD/configs/person.jpg -------------------------------------------------------------------------------- /configs/result.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/enazoe/yolo-tensorrt/HEAD/configs/result.jpg -------------------------------------------------------------------------------- /configs/qq_group.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/enazoe/yolo-tensorrt/HEAD/configs/qq_group.png -------------------------------------------------------------------------------- /configs/qrcode.jpeg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/enazoe/yolo-tensorrt/HEAD/configs/qrcode.jpeg -------------------------------------------------------------------------------- /configs/yolo-trt.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/enazoe/yolo-tensorrt/HEAD/configs/yolo-trt.png -------------------------------------------------------------------------------- /modules/decodeTensorCUDA.h: -------------------------------------------------------------------------------- 1 | #ifndef DECODETENSORCUDA_H_ 2 | #define DECODETENSORCUDA_H_ 3 | #include "yolo.h" 4 | 5 | float* decodeTensorCUDA(const int imageIdx, const TensorInfo& tensor); 6 | #endif -------------------------------------------------------------------------------- /extra/API.h: -------------------------------------------------------------------------------- 1 | #ifdef API_EXPORTS 2 | 3 | #if defined(_MSC_VER) 4 | #define API __declspec(dllexport) 5 | #else 6 | #define API __attribute__((visibility("default"))) 7 | #endif 8 | 9 | #else 10 | 11 | #if defined(_MSC_VER) 12 | #define API __declspec(dllimport) 13 | #else 14 | #define API 15 | #endif 16 | 17 | #endif -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- 1 | build/ 2 | *.weights 3 | *.engine 4 | *.pdb 5 | x64/ 6 | ipch/ 7 | data/ 8 | *.sdf 9 | *.opensdf 10 | *.exp 11 | *.db 12 | *.ipdb 13 | *.dll 14 | *.lib 15 | *.table 16 | api/ 17 | *.iobj 18 | *.opendb 19 | *.exe 20 | .vs/ 21 | .vscode/ 22 | *.log 23 | *.pt 24 | *.onnx 25 | .ipynb_checkpoints 26 | *.idea 27 | Dockerfile 28 | -------------------------------------------------------------------------------- /sln/test_dll/test_dll.vcxproj.user: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | $(SolutionDir) 5 | WindowsLocalDebugger 6 | 7 | -------------------------------------------------------------------------------- /modules/yolov4.h: -------------------------------------------------------------------------------- 1 | #ifndef CLASS_YOLOV4_H_ 2 | #define CLASS_YOLOV4_H_ 3 | #include "yolo.h" 4 | class YoloV4 :public Yolo 5 | { 6 | public: 7 | YoloV4( 8 | const NetworkInfo &network_info_, 9 | const InferParams &infer_params_); 10 | private: 11 | std::vector decodeTensor(const int imageIdx, 12 | const int imageH, 13 | const int imageW, 14 | const TensorInfo& tensor) override; 15 | }; 16 | 17 | #endif 18 | -------------------------------------------------------------------------------- /sln/dll_detector/dll_detector.vcxproj.user: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | $(SolutionDir) 5 | WindowsLocalDebugger 6 | 7 | -------------------------------------------------------------------------------- /modules/class_detector.cpp: -------------------------------------------------------------------------------- 1 | #include "class_detector.h" 2 | #include "class_yolo_detector.hpp" 3 | 4 | class Detector::Impl 5 | { 6 | public: 7 | Impl() {} 8 | 9 | ~Impl(){} 10 | 11 | YoloDectector _detector; 12 | }; 13 | 14 | Detector::Detector() 15 | { 16 | _impl = new Impl(); 17 | } 18 | 19 | Detector::~Detector() 20 | { 21 | if (_impl) 22 | { 23 | delete _impl; 24 | _impl = nullptr; 25 | } 26 | } 27 | 28 | void Detector::init(const Config &config) 29 | { 30 | _impl->_detector.init(config); 31 | } 32 | 33 | void Detector::detect(const std::vector &mat_image, std::vector &vec_batch_result) 34 | { 35 | _impl->_detector.detect(mat_image, vec_batch_result); 36 | } 37 | -------------------------------------------------------------------------------- /sln/props/opencv412_release_x64.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | D:\dependency\OPENCV412\include;%(AdditionalIncludeDirectories) 9 | 10 | 11 | D:\dependency\OPENCV412\x64\vc14\lib;%(AdditionalLibraryDirectories) 12 | opencv_world412.lib;%(AdditionalDependencies) 13 | 14 | 15 | 16 | -------------------------------------------------------------------------------- /extra/class_timer.hpp: -------------------------------------------------------------------------------- 1 | #pragma once 2 | 3 | #include 4 | #include 5 | #include 6 | 7 | class Timer 8 | { 9 | public: 10 | Timer() : beg_(clock_::now()) 11 | {} 12 | void reset() 13 | { 14 | beg_ = clock_::now(); 15 | } 16 | 17 | double elapsed() const 18 | { 19 | return std::chrono::duration_cast(clock_::now() - beg_).count(); 20 | } 21 | 22 | void out(std::string message = "") 23 | { 24 | double t = elapsed(); 25 | std::cout << message << " elasped time:" << t << "ms" << std::endl; 26 | reset(); 27 | } 28 | 29 | double get_duration()const 30 | { 31 | return elapsed(); 32 | } 33 | private: 34 | using clock_ = std::chrono::high_resolution_clock; 35 | using second_ = std::chrono::duration; 36 | std::chrono::time_point beg_; 37 | }; 38 | -------------------------------------------------------------------------------- /sln/props/TensorRT-8.0.1.6_cuda-11.3_cudnn8.2.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | D:\dependency\TensorRT-8.0.1.6_cuda-11.3_cudnn8.2\include;%(AdditionalIncludeDirectories) 9 | 10 | 11 | D:\dependency\TensorRT-8.0.1.6_cuda-11.3_cudnn8.2\lib;%(AdditionalLibraryDirectories) 12 | nvinfer.lib;%(AdditionalDependencies) 13 | 14 | 15 | 16 | -------------------------------------------------------------------------------- /sln/props/cuda100_x64_release.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\include;%(AdditionalIncludeDirectories) 9 | 10 | 11 | C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\lib\x64;%(AdditionalLibraryDirectories) 12 | cudart_static.lib;%(AdditionalDependencies) 13 | 14 | 15 | 16 | -------------------------------------------------------------------------------- /sln/props/cuda11_x64_release.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.0\include;%(AdditionalIncludeDirectories) 9 | 10 | 11 | C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.0\lib\x64;%(AdditionalLibraryDirectories) 12 | cudart_static.lib;%(AdditionalDependencies) 13 | 14 | 15 | 16 | -------------------------------------------------------------------------------- /scripts/CMakeLists.txt: -------------------------------------------------------------------------------- 1 | cmake_minimum_required(VERSION 3.1) # setting this is required 2 | project(yolov3-trt) # this sets the project name 3 | 4 | 5 | 6 | add_executable(yolov3-trt samples/sample_detector.cpp) 7 | 8 | 9 | set(CMAKE_CXX_COMPILIER "/usr/bin/g++") 10 | set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wno-write-strings") 11 | 12 | LINK_DIRECTORIES(${CMAKE_BINARY_DIR}) 13 | message(STATUS "${CMAKE_BINARY_DIR}") 14 | message(STATUS "${CMAKE_SOURCE_DIR}") 15 | 16 | add_library(detector SHARED IMPORTED) 17 | set_target_properties(detector PROPERTIES IMPORTED_LOCATION ${CMAKE_SOURCE_DIR}/lib/libdetector.so) 18 | 19 | find_package(OpenCV REQUIRED) 20 | 21 | 22 | target_include_directories(yolov3-trt PRIVATE modules/ ./) 23 | 24 | target_link_libraries(yolov3-trt detector opencv_core opencv_calib3d opencv_dnn) 25 | 26 | 27 | 28 | 29 | 30 | -------------------------------------------------------------------------------- /scripts/download.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env python3 2 | 3 | import requests 4 | print("download cfg file from https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov3.cfg ...") 5 | url = 'https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov3.cfg' 6 | r = requests.get(url) 7 | 8 | with open('../configs/yolov3.cfg', 'wb') as f: 9 | f.write(r.content) 10 | print("done!") 11 | print("download weights file from https://pjreddie.com/media/files/yolov3.weights ...") 12 | 13 | url = 'https://pjreddie.com/media/files/yolov3.weights' 14 | r = requests.get(url) 15 | 16 | with open('../configs/yolov3.weights', 'wb') as f: 17 | f.write(r.content) 18 | print("done!") 19 | 20 | url = 'https://raw.githubusercontent.com/pjreddie/darknet/master/data/dog.jpg' 21 | r = requests.get(url) 22 | 23 | with open('../configs/dog.jpg', 'wb') as f: 24 | f.write(r.content) 25 | print("complete !") -------------------------------------------------------------------------------- /sln/props/gflags_x64_release.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | <_PropertySheetDisplayName>gflag_x64_release 7 | 8 | 9 | 10 | D:\dependency\gflag\include;%(AdditionalIncludeDirectories) 11 | 12 | 13 | D:\dependency\gflag\lib;%(AdditionalLibraryDirectories) 14 | gflags_static.lib;gflags.lib;%(AdditionalDependencies) 15 | 16 | 17 | 18 | -------------------------------------------------------------------------------- /sln/props/TensorRT-5.1.5.0_cuda-10.0_cudnn-7.5.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | D:\dependency\TensorRT-5.1.5.0_cuda-10.0_cudnn-7.5\include;%(AdditionalIncludeDirectories) 9 | 10 | 11 | D:\dependency\TensorRT-5.1.5.0_cuda-10.0_cudnn-7.5\lib;%(AdditionalLibraryDirectories) 12 | nvinfer.lib;nvinfer_plugin.lib;nvonnxparser.lib;nvparsers.lib;%(AdditionalDependencies) 13 | 14 | 15 | 16 | -------------------------------------------------------------------------------- /sln/props/opencv330_release_x64.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | <_PropertySheetDisplayName>opencv330_release_x64 7 | 8 | 9 | 10 | D:\dependency\OPENCV330\include;%(AdditionalIncludeDirectories) 11 | 12 | 13 | D:\dependency\OPENCV330\x64\vc12\lib;%(AdditionalLibraryDirectories) 14 | opencv_world330.lib;%(AdditionalDependencies) 15 | 16 | 17 | 18 | -------------------------------------------------------------------------------- /sln/props/TensorRT-7.1.3.4_cuda-11.0_cudnn8.0.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | D:\dependency\TensorRT-7.1.3.4_cuda-11.0.cudnn8.0\include;%(AdditionalIncludeDirectories) 9 | 10 | 11 | D:\dependency\TensorRT-7.1.3.4_cuda-11.0.cudnn8.0\lib;%(AdditionalLibraryDirectories) 12 | nvinfer.lib;nvinfer_plugin.lib;nvonnxparser.lib;nvparsers.lib;myelin64_1.lib;%(AdditionalDependencies) 13 | 14 | 15 | 16 | -------------------------------------------------------------------------------- /sln/props/TensorRT-7.2.1.6_cuda-11.0.cudnn8.0.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | D:\dependency\TensorRT-7.2.1.6_cuda-11.0.cudnn8.0\include;%(AdditionalIncludeDirectories) 9 | 10 | 11 | D:\dependency\TensorRT-7.2.1.6_cuda-11.0.cudnn8.0\lib;%(AdditionalLibraryDirectories) 12 | myelin64_1.lib;nvinfer.lib;nvinfer_plugin.lib;nvonnxparser.lib;nvparsers.lib;%(AdditionalDependencies) 13 | 14 | 15 | 16 | -------------------------------------------------------------------------------- /sln/props/TensorRT-7.2.3.4_cuda-11.1.cudnn8.1.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | D:\dependency\TensorRT-7.2.3.4.Windows10.x86_64.cuda-11.1.cudnn8.1\include;%(AdditionalIncludeDirectories) 9 | 10 | 11 | D:\dependency\TensorRT-7.2.3.4.Windows10.x86_64.cuda-11.1.cudnn8.1\lib;%(AdditionalLibraryDirectories) 12 | myelin64_1.lib;nvinfer.lib;nvinfer_plugin.lib;nvonnxparser.lib;nvparsers.lib;%(AdditionalDependencies) 13 | 14 | 15 | 16 | -------------------------------------------------------------------------------- /sln/dll_detector/TensorRT-8.0.1.6_cuda-10.2.cudnn8.2.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | D:\dependency\TensorRT-8.0.1.6\include;%(AdditionalIncludeDirectories) 9 | 10 | 11 | D:\dependency\TensorRT-8.0.1.6\lib;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\lib\x64;%(AdditionalLibraryDirectories) 12 | nvinfer.lib;nvinfer_plugin.lib;nvonnxparser.lib;nvparsers.lib;cudart.lib;cuda.lib;cudnn.lib;cublas.lib;%(AdditionalDependencies) 13 | 14 | 15 | 16 | -------------------------------------------------------------------------------- /sln/props/dll_debug_release_x64.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | <_PropertySheetDisplayName>linx_dll_debug_release_x64 7 | $(SolutionDir) 8 | $(SolutionDir)$(Platform)\ 9 | 10 | 11 | 12 | ..\..\modules;..\..\extra;%(AdditionalIncludeDirectories) 13 | API_EXPORTS;%(PreprocessorDefinitions) 14 | 15 | 16 | $(SolutionDir)$(Configuration)\;%(AdditionalLibraryDirectories) 17 | 18 | 19 | 20 | -------------------------------------------------------------------------------- /modules/yolov5.h: -------------------------------------------------------------------------------- 1 | #ifndef CLASS_YOLOV5_H_ 2 | #define CLASS_YOLOV5_H_ 3 | #include "yolo.h" 4 | class YoloV5 :public Yolo 5 | { 6 | public: 7 | YoloV5( 8 | const NetworkInfo &network_info_, 9 | const InferParams &infer_params_); 10 | 11 | BBox convert_bbox_res(const float& bx, const float& by, const float& bw, const float& bh, 12 | const uint32_t& stride_h_, const uint32_t& stride_w_, const uint32_t& netW, const uint32_t& netH) 13 | { 14 | BBox b; 15 | // Restore coordinates to network input resolution 16 | float x = bx * stride_w_; 17 | float y = by * stride_h_; 18 | 19 | b.x1 = x - bw / 2; 20 | b.x2 = x + bw / 2; 21 | 22 | b.y1 = y - bh / 2; 23 | b.y2 = y + bh / 2; 24 | 25 | b.x1 = clamp(b.x1, 0, netW); 26 | b.x2 = clamp(b.x2, 0, netW); 27 | b.y1 = clamp(b.y1, 0, netH); 28 | b.y2 = clamp(b.y2, 0, netH); 29 | 30 | return b; 31 | } 32 | 33 | 34 | private: 35 | std::vector decodeTensor(const int imageIdx, 36 | const int imageH, 37 | const int imageW, 38 | const TensorInfo& tensor) override; 39 | }; 40 | 41 | #endif -------------------------------------------------------------------------------- /sln/props/console_debug_release_x64 .props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | <_PropertySheetDisplayName>linx_console_debug_release_x64 7 | $(SolutionDir) 8 | $(SolutionDir)$(Platform) 9 | 10 | 11 | 12 | ..\..\modules;..\..\extra\linx;..\..\extra\linx_timer;%(AdditionalIncludeDirectories) 13 | _CRT_SECURE_NO_WARNINGS;%(PreprocessorDefinitions) 14 | 15 | 16 | $(SolutionDir)$(Configuration)\;..\..\libs;$(SolutionDir);%(AdditionalLibraryDirectories) 17 | 18 | 19 | 20 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2020 enazoe 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. 22 | -------------------------------------------------------------------------------- /CMakeLists.txt: -------------------------------------------------------------------------------- 1 | cmake_minimum_required(VERSION 3.1) 2 | project(yolo-trt VERSION 1.0) 3 | enable_language(CUDA) 4 | 5 | set(CMAKE_CXX_COMPILIER "/usr/bin/g++") 6 | set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14 -O3 -Wno-write-strings") 7 | set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,-rpath -Wl,$ORIGIN") 8 | 9 | #cuda 10 | #export PATH=/usr/local/cuda-11.0/bin:$PATH 11 | #include_directories(/usr/local/cuda/include) 12 | #link_directories(/usr/local/cuda/lib64) 13 | find_package(CUDA REQUIRED) 14 | 15 | #opencv 16 | #set(OpenCV_DIR /opt/opencv-4.1.2/share/OpenCV) 17 | find_package(OpenCV REQUIRED) 18 | 19 | #detector lib (cuda) 20 | file(GLOB_RECURSE sources modules/*.hpp modules/*.cpp modules/*.h modules/*.cu extra/*.h) 21 | add_library(detector SHARED ${sources}) 22 | target_include_directories(detector PRIVATE extra/ modules/ ${OpenCV_INCLUDE_DIRS} ${CUDA_TOOLKIT_ROOT_DIR}/include) 23 | target_link_libraries(detector nvinfer nvinfer_plugin nvparsers "stdc++fs") 24 | 25 | #sample 26 | add_executable(yolo-trt samples/sample_detector.cpp) 27 | target_include_directories(yolo-trt PRIVATE modules/ extra/) 28 | target_link_libraries(yolo-trt detector ${OpenCV_LIBS}) 29 | 30 | 31 | 32 | -------------------------------------------------------------------------------- /sln/test_dll/test_dll.vcxproj.filters: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | {4FC737F1-C7A5-4376-A066-2A32D752A2FF} 6 | cpp;c;cc;cxx;def;odl;idl;hpj;bat;asm;asmx 7 | 8 | 9 | {93995380-89BD-4b04-88EB-625FBE52EBFB} 10 | h;hh;hpp;hxx;hm;inl;inc;xsd 11 | 12 | 13 | {67DA6AB6-F800-4c08-8B7A-83BB121AAD01} 14 | rc;ico;cur;bmp;dlg;rc2;rct;bin;rgs;gif;jpg;jpeg;jpe;resx;tiff;tif;png;wav;mfcribbon-ms 15 | 16 | 17 | 18 | 19 | Source Files 20 | 21 | 22 | 23 | 24 | Header Files 25 | 26 | 27 | -------------------------------------------------------------------------------- /modules/class_detector.h: -------------------------------------------------------------------------------- 1 | #ifndef CLASS_DETECTOR_H_ 2 | #define CLASS_DETECTOR_H_ 3 | 4 | #include "API.h" 5 | #include 6 | #include 7 | 8 | struct Result 9 | { 10 | int id = -1; 11 | float prob = 0.f; 12 | cv::Rect rect; 13 | }; 14 | 15 | using BatchResult = std::vector; 16 | 17 | enum ModelType 18 | { 19 | YOLOV3, 20 | YOLOV4, 21 | YOLOV5 22 | }; 23 | 24 | enum Precision 25 | { 26 | INT8 = 0, 27 | FP16, 28 | FP32 29 | }; 30 | 31 | struct Config 32 | { 33 | std::string file_model_cfg = "configs/yolov4.cfg"; 34 | 35 | std::string file_model_weights = "configs/yolov4.weights"; 36 | 37 | float detect_thresh = 0.9; 38 | 39 | ModelType net_type = YOLOV4; 40 | 41 | Precision inference_precison = FP32; 42 | 43 | int gpu_id = 0; 44 | 45 | std::string calibration_image_list_file_txt = "configs/calibration_images.txt"; 46 | 47 | }; 48 | 49 | class API Detector 50 | { 51 | public: 52 | explicit Detector(); 53 | 54 | ~Detector(); 55 | 56 | void init(const Config &config); 57 | 58 | void detect(const std::vector &mat_image, std::vector &vec_batch_result); 59 | 60 | private: 61 | 62 | Detector(const Detector &); 63 | const Detector &operator =(const Detector &); 64 | class Impl; 65 | Impl *_impl; 66 | }; 67 | 68 | #endif // !CLASS_QH_DETECTOR_H_ -------------------------------------------------------------------------------- /sln/props/opencv412_release_x64_vs2017.props: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | D:\dependency\OpenCV412\x64\vc15\bin;D:\dependency\OpenCV412\include;$(VC_IncludePath);$(WindowsSDK_IncludePath);$(IncludePath) 7 | D:\dependency\OpenCV412\x64\vc15\lib;$(VC_LibraryPath_x64);$(WindowsSDK_LibraryPath_x64);$(NETFXKitsDir)Lib\um\x64;$(LibraryPath) 8 | $(VC_ExecutablePath_x64);$(WindowsSDK_ExecutablePath);$(VS_ExecutablePath);$(MSBuild_ExecutablePath);$(FxCopDir);$(PATH); 9 | <_PropertySheetDisplayName>OpenCV412_release_x64_vs2017 10 | 11 | 12 | 13 | opencv_world412.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies) 14 | D:\dependency\OpenCV412\x64\vc15\lib;%(AdditionalLibraryDirectories) 15 | 16 | 17 | 18 | 64 19 | 20 | 21 | 22 | -------------------------------------------------------------------------------- /configs/yolov5-4.0/yolov5l.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5l.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-4.0/yolov5m.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.67 # model depth multiple 4 | width_multiple: 0.75 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-4.0/yolov5s.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-4.0/yolov5x.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5m.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.67 # model depth multiple 4 | width_multiple: 0.75 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5s.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5x.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, C3, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5l.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.0 # model depth multiple 6 | width_multiple: 1.0 # layer channel multiple 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5m.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.67 # model depth multiple 6 | width_multiple: 0.75 # layer channel multiple 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5n.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.33 # model depth multiple 6 | width_multiple: 0.25 # layer channel multiple 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5s.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 0.33 # model depth multiple 6 | width_multiple: 0.50 # layer channel multiple 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5x.yaml: -------------------------------------------------------------------------------- 1 | # YOLOv5 🚀 by Ultralytics, GPL-3.0 license 2 | 3 | # Parameters 4 | nc: 80 # number of classes 5 | depth_multiple: 1.33 # model depth multiple 6 | width_multiple: 1.25 # layer channel multiple 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 v6.0 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Conv, [64, 6, 2, 2]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, C3, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 6, C3, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, C3, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 3, C3, [1024]], 24 | [-1, 1, SPPF, [1024, 5]], # 9 25 | ] 26 | 27 | # YOLOv5 v6.0 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, C3, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, C3, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, C3, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, C3, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-2.0/yolov5l.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-2.0/yolov5m.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.67 # model depth multiple 4 | width_multiple: 0.75 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-2.0/yolov5s.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-2.0/yolov5x.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-3.0/yolov5l.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-3.0/yolov5m.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.67 # model depth multiple 4 | width_multiple: 0.75 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-3.0/yolov5s.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /configs/yolov5-3.0/yolov5x.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [10,13, 16,30, 33,23] # P3/8 9 | - [30,61, 62,45, 59,119] # P4/16 10 | - [116,90, 156,198, 373,326] # P5/32 11 | 12 | # YOLOv5 backbone 13 | backbone: 14 | # [from, number, module, args] 15 | [[-1, 1, Focus, [64, 3]], # 0-P1/2 16 | [-1, 1, Conv, [128, 3, 2]], # 1-P2/4 17 | [-1, 3, BottleneckCSP, [128]], 18 | [-1, 1, Conv, [256, 3, 2]], # 3-P3/8 19 | [-1, 9, BottleneckCSP, [256]], 20 | [-1, 1, Conv, [512, 3, 2]], # 5-P4/16 21 | [-1, 9, BottleneckCSP, [512]], 22 | [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32 23 | [-1, 1, SPP, [1024, [5, 9, 13]]], 24 | [-1, 3, BottleneckCSP, [1024, False]], # 9 25 | ] 26 | 27 | # YOLOv5 head 28 | head: 29 | [[-1, 1, Conv, [512, 1, 1]], 30 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 31 | [[-1, 6], 1, Concat, [1]], # cat backbone P4 32 | [-1, 3, BottleneckCSP, [512, False]], # 13 33 | 34 | [-1, 1, Conv, [256, 1, 1]], 35 | [-1, 1, nn.Upsample, [None, 2, 'nearest']], 36 | [[-1, 4], 1, Concat, [1]], # cat backbone P3 37 | [-1, 3, BottleneckCSP, [256, False]], # 17 (P3/8-small) 38 | 39 | [-1, 1, Conv, [256, 3, 2]], 40 | [[-1, 14], 1, Concat, [1]], # cat head P4 41 | [-1, 3, BottleneckCSP, [512, False]], # 20 (P4/16-medium) 42 | 43 | [-1, 1, Conv, [512, 3, 2]], 44 | [[-1, 10], 1, Concat, [1]], # cat head P5 45 | [-1, 3, BottleneckCSP, [1024, False]], # 23 (P5/32-large) 46 | 47 | [[17, 20, 23], 1, Detect, [nc, anchors]], # Detect(P3, P4, P5) 48 | ] 49 | -------------------------------------------------------------------------------- /modules/yolov5.cpp: -------------------------------------------------------------------------------- 1 | 2 | #include "yolov5.h" 3 | #include "decodeTensorCUDA.h" 4 | 5 | YoloV5::YoloV5( 6 | const NetworkInfo &network_info_, 7 | const InferParams &infer_params_) : 8 | Yolo( network_info_, infer_params_) {} 9 | std::vector YoloV5::decodeTensor(const int imageIdx, const int imageH, const int imageW, const TensorInfo& tensor) 10 | { 11 | float scale_h = 1.f; 12 | float scale_w = 1.f; 13 | int xOffset = 0; 14 | int yOffset = 0; 15 | calcuate_letterbox_message(m_InputH, m_InputW, imageH, imageW, scale_h, scale_w, xOffset, yOffset); 16 | std::vector binfo; 17 | 18 | // decode output (x,y,w,h,maxProb,maxIndex) 19 | float* boxes = decodeTensorCUDA(imageIdx, tensor); 20 | 21 | // Ergodic result 22 | for (uint32_t y = 0; y < tensor.grid_h; ++y) 23 | { 24 | for (uint32_t x = 0; x < tensor.grid_w; ++x) 25 | { 26 | for (uint32_t b = 0; b < tensor.numBBoxes; ++b) 27 | { 28 | const int bbindex = y * tensor.grid_w+ x; 29 | const float bx 30 | = boxes[18 * bbindex + 6*b + 0]; 31 | 32 | const float by 33 | = boxes[18 * bbindex + 6*b + 1]; 34 | 35 | const float bw 36 | = boxes[18 * bbindex + 6*b + 2]; 37 | 38 | const float bh 39 | = boxes[18 * bbindex + 6*b + 3]; 40 | 41 | const float maxProb 42 | = boxes[18 * bbindex + 6*b + 4]; 43 | 44 | const int maxIndex 45 | = (int) boxes[18 * bbindex + 6*b + 5]; 46 | 47 | if (maxProb > m_ProbThresh) 48 | { 49 | add_bbox_proposal(bx, by, bw, bh, tensor.stride_h, tensor.stride_w, scale_h, scale_w, xOffset, yOffset, maxIndex, maxProb, imageW, imageH, binfo); 50 | } 51 | 52 | } 53 | } 54 | } 55 | return binfo; 56 | } 57 | -------------------------------------------------------------------------------- /modules/yolov3.h: -------------------------------------------------------------------------------- 1 | /** 2 | MIT License 3 | 4 | Copyright (c) 2018 NVIDIA CORPORATION. All rights reserved. 5 | 6 | Permission is hereby granted, free of charge, to any person obtaining a copy 7 | of this software and associated documentation files (the "Software"), to deal 8 | in the Software without restriction, including without limitation the rights 9 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 | copies of the Software, and to permit persons to whom the Software is 11 | furnished to do so, subject to the following conditions: 12 | 13 | The above copyright notice and this permission notice shall be included in all 14 | copies or substantial portions of the Software. 15 | 16 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 | SOFTWARE. 23 | * 24 | */ 25 | 26 | #ifndef _YOLO_V3_ 27 | #define _YOLO_V3_ 28 | 29 | #include "yolo.h" 30 | 31 | #include 32 | #include 33 | #include 34 | 35 | class YoloV3 : public Yolo 36 | { 37 | public: 38 | YoloV3(const NetworkInfo& networkInfo, const InferParams& inferParams); 39 | 40 | private: 41 | std::vector decodeTensor(const int imageIdx, const int imageH, const int imageW, 42 | const TensorInfo& tensor) override; 43 | }; 44 | 45 | #endif // _YOLO_V3_ -------------------------------------------------------------------------------- /modules/yolov2.h: -------------------------------------------------------------------------------- 1 | /** 2 | MIT License 3 | 4 | Copyright (c) 2018 NVIDIA CORPORATION. All rights reserved. 5 | 6 | Permission is hereby granted, free of charge, to any person obtaining a copy 7 | of this software and associated documentation files (the "Software"), to deal 8 | in the Software without restriction, including without limitation the rights 9 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 | copies of the Software, and to permit persons to whom the Software is 11 | furnished to do so, subject to the following conditions: 12 | 13 | The above copyright notice and this permission notice shall be included in all 14 | copies or substantial portions of the Software. 15 | 16 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 | SOFTWARE. 23 | * 24 | */ 25 | 26 | #ifndef _YOLO_V2_ 27 | #define _YOLO_V2_ 28 | 29 | #include "yolo.h" 30 | 31 | #include 32 | #include 33 | #include 34 | 35 | class YoloV2 : public Yolo 36 | { 37 | public: 38 | YoloV2( const NetworkInfo& networkInfo, const InferParams& inferParams); 39 | 40 | private: 41 | std::vector decodeTensor(const int imageIdx, const int imageH, const int imageW, 42 | const TensorInfo& tensor) override; 43 | }; 44 | 45 | #endif // _YOLO_V2_ -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5l.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.0 9 | width_multiple=1.0 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Conv] 13 | from=-1 14 | number=1 15 | args=64, 6, 2, 2 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=6 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [C3] 53 | from=-1 54 | number=3 55 | args=1024 56 | 57 | [SPPF] 58 | from=-1 59 | number=1 60 | args=1024, 5 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5s.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=0.33 9 | width_multiple=0.5 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Conv] 13 | from=-1 14 | number=1 15 | args=64, 6, 2, 2 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=6 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [C3] 53 | from=-1 54 | number=3 55 | args=1024 56 | 57 | [SPPF] 58 | from=-1 59 | number=1 60 | args=1024, 5 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5m.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=0.67 9 | width_multiple=0.75 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Conv] 13 | from=-1 14 | number=1 15 | args=64, 6, 2, 2 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=6 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [C3] 53 | from=-1 54 | number=3 55 | args=1024 56 | 57 | [SPPF] 58 | from=-1 59 | number=1 60 | args=1024, 5 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5n.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=0.33 9 | width_multiple=0.25 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Conv] 13 | from=-1 14 | number=1 15 | args=64, 6, 2, 2 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=6 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [C3] 53 | from=-1 54 | number=3 55 | args=1024 56 | 57 | [SPPF] 58 | from=-1 59 | number=1 60 | args=1024, 5 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-6.0/yolov5x.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.33 9 | width_multiple=1.25 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Conv] 13 | from=-1 14 | number=1 15 | args=64, 6, 2, 2 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=6 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [C3] 53 | from=-1 54 | number=3 55 | args=1024 56 | 57 | [SPPF] 58 | from=-1 59 | number=1 60 | args=1024, 5 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-4.0/yolov5l.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.0 9 | width_multiple=1.0 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [C3] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5l.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.0 9 | width_multiple=1.0 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [C3] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-4.0/yolov5m.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=0.67 9 | width_multiple=0.75 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [C3] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-4.0/yolov5s.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch= 2 7 | nc=80 8 | depth_multiple=0.33 9 | width_multiple=0.5 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [C3] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-4.0/yolov5x.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.33 9 | width_multiple=1.25 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [C3] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5m.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=0.67 9 | width_multiple=0.75 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [C3] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5s.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch= 1 7 | nc=80 8 | depth_multiple=0.33 9 | width_multiple=0.5 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [C3] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5x.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.33 9 | width_multiple=1.25 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [C3] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [C3] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [C3] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [C3] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [C3] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [C3] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [C3] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [C3] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /modules/yolo_config_parser.h: -------------------------------------------------------------------------------- 1 | /** 2 | MIT License 3 | 4 | Copyright (c) 2018 NVIDIA CORPORATION. All rights reserved. 5 | 6 | Permission is hereby granted, free of charge, to any person obtaining a copy 7 | of this software and associated documentation files (the "Software"), to deal 8 | in the Software without restriction, including without limitation the rights 9 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 | copies of the Software, and to permit persons to whom the Software is 11 | furnished to do so, subject to the following conditions: 12 | 13 | The above copyright notice and this permission notice shall be included in all 14 | copies or substantial portions of the Software. 15 | 16 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 | SOFTWARE. 23 | * 24 | */ 25 | 26 | #ifndef _YOLO_CONFIG_PARSER_ 27 | #define _YOLO_CONFIG_PARSER_ 28 | 29 | #include "yolo.h" 30 | 31 | #include 32 | //#include 33 | 34 | // Init to be called at the very beginning to verify all config params are valid 35 | //void yoloConfigParserInit(int argc, char** argv); 36 | 37 | //NetworkInfo getYoloNetworkInfo(); 38 | //InferParams getYoloInferParams(); 39 | //uint64_t getSeed(); 40 | //std::string getNetworkType(); 41 | //std::string getPrecision(); 42 | //std::string getTestImages(); 43 | //std::string getTestImagesPath(); 44 | //bool getDecode(); 45 | //bool getDoBenchmark(); 46 | //bool getViewDetections(); 47 | //bool getSaveDetections(); 48 | //std::string getSaveDetectionsPath(); 49 | //uint32_t getBatchSize(); 50 | //bool getShuffleTestSet(); 51 | 52 | #endif //_YOLO_CONFIG_PARSER_ -------------------------------------------------------------------------------- /configs/yolov5-2.0/yolov5m.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=320 5 | channels=3 6 | nc=80 7 | depth_multiple=0.67 8 | width_multiple=0.75 9 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 10 | 11 | [Focus] 12 | from=-1 13 | number=1 14 | args=64, 3 15 | 16 | [Conv] 17 | from=-1 18 | number=1 19 | args=128, 3, 2 20 | 21 | [BottleneckCSP] 22 | from=-1 23 | number=3 24 | args=128 25 | 26 | [Conv] 27 | from=-1 28 | number=1 29 | args=256, 3, 2 30 | 31 | [BottleneckCSP] 32 | from=-1 33 | number=9 34 | args=256 35 | 36 | [Conv] 37 | from=-1 38 | number=1 39 | args=512, 3, 2 40 | 41 | [BottleneckCSP] 42 | from=-1 43 | number=9 44 | args=512 45 | 46 | [Conv] 47 | from=-1 48 | number=1 49 | args=1024, 3, 2 50 | 51 | [SPP] 52 | from=-1 53 | number=1 54 | args=1024, [5, 9, 13] 55 | 56 | [BottleneckCSP] 57 | from=-1 58 | number=3 59 | args=1024, False 60 | 61 | [Conv] 62 | from=-1 63 | number=1 64 | args=512, 1, 1 65 | 66 | [nn.Upsample] 67 | from=-1 68 | number=1 69 | args='None', 2, 'nearest' 70 | 71 | [Concat] 72 | from=-1, 6 73 | number=1 74 | args=1 75 | 76 | [BottleneckCSP] 77 | from=-1 78 | number=3 79 | args=512, False 80 | 81 | [Conv] 82 | from=-1 83 | number=1 84 | args=256, 1, 1 85 | 86 | [nn.Upsample] 87 | from=-1 88 | number=1 89 | args='None', 2, 'nearest' 90 | 91 | [Concat] 92 | from=-1, 4 93 | number=1 94 | args=1 95 | 96 | [BottleneckCSP] 97 | from=-1 98 | number=3 99 | args=256, False 100 | 101 | [Conv] 102 | from=-1 103 | number=1 104 | args=256, 3, 2 105 | 106 | [Concat] 107 | from=-1, 14 108 | number=1 109 | args=1 110 | 111 | [BottleneckCSP] 112 | from=-1 113 | number=3 114 | args=512, False 115 | 116 | [Conv] 117 | from=-1 118 | number=1 119 | args=512, 3, 2 120 | 121 | [Concat] 122 | from=-1, 10 123 | number=1 124 | args=1 125 | 126 | [BottleneckCSP] 127 | from=-1 128 | number=3 129 | args=1024, False 130 | 131 | [Detect] 132 | from=17, 20, 23 133 | number=1 134 | args='nc', 'anchors' 135 | -------------------------------------------------------------------------------- /configs/yolov5-2.0/yolov5l.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=320 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.0 9 | width_multiple=1.0 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [BottleneckCSP] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [BottleneckCSP] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [BottleneckCSP] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [BottleneckCSP] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [BottleneckCSP] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [BottleneckCSP] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [BottleneckCSP] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [BottleneckCSP] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-2.0/yolov5s.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=320 4 | height=320 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=0.33 9 | width_multiple=0.5 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [BottleneckCSP] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [BottleneckCSP] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [BottleneckCSP] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [BottleneckCSP] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [BottleneckCSP] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [BottleneckCSP] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [BottleneckCSP] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [BottleneckCSP] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-2.0/yolov5x.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=320 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.33 9 | width_multiple=1.25 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [BottleneckCSP] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [BottleneckCSP] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [BottleneckCSP] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [BottleneckCSP] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [BottleneckCSP] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [BottleneckCSP] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [BottleneckCSP] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [BottleneckCSP] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-3.0/yolov5l.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.0 9 | width_multiple=1.0 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [BottleneckCSP] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [BottleneckCSP] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [BottleneckCSP] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [BottleneckCSP] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [BottleneckCSP] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [BottleneckCSP] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [BottleneckCSP] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [BottleneckCSP] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-3.0/yolov5m.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=0.67 9 | width_multiple=0.75 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [BottleneckCSP] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [BottleneckCSP] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [BottleneckCSP] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [BottleneckCSP] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [BottleneckCSP] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [BottleneckCSP] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [BottleneckCSP] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [BottleneckCSP] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-3.0/yolov5s.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=0.33 9 | width_multiple=0.5 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [BottleneckCSP] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [BottleneckCSP] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [BottleneckCSP] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [BottleneckCSP] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [BottleneckCSP] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [BottleneckCSP] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [BottleneckCSP] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [BottleneckCSP] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /configs/yolov5-3.0/yolov5x.cfg: -------------------------------------------------------------------------------- 1 | 2 | [net] 3 | width=640 4 | height=640 5 | channels=3 6 | batch=1 7 | nc=80 8 | depth_multiple=1.33 9 | width_multiple=1.25 10 | anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 11 | 12 | [Focus] 13 | from=-1 14 | number=1 15 | args=64, 3 16 | 17 | [Conv] 18 | from=-1 19 | number=1 20 | args=128, 3, 2 21 | 22 | [BottleneckCSP] 23 | from=-1 24 | number=3 25 | args=128 26 | 27 | [Conv] 28 | from=-1 29 | number=1 30 | args=256, 3, 2 31 | 32 | [BottleneckCSP] 33 | from=-1 34 | number=9 35 | args=256 36 | 37 | [Conv] 38 | from=-1 39 | number=1 40 | args=512, 3, 2 41 | 42 | [BottleneckCSP] 43 | from=-1 44 | number=9 45 | args=512 46 | 47 | [Conv] 48 | from=-1 49 | number=1 50 | args=1024, 3, 2 51 | 52 | [SPP] 53 | from=-1 54 | number=1 55 | args=1024, [5, 9, 13] 56 | 57 | [BottleneckCSP] 58 | from=-1 59 | number=3 60 | args=1024, False 61 | 62 | [Conv] 63 | from=-1 64 | number=1 65 | args=512, 1, 1 66 | 67 | [nn.Upsample] 68 | from=-1 69 | number=1 70 | args='None', 2, 'nearest' 71 | 72 | [Concat] 73 | from=-1, 6 74 | number=1 75 | args=1 76 | 77 | [BottleneckCSP] 78 | from=-1 79 | number=3 80 | args=512, False 81 | 82 | [Conv] 83 | from=-1 84 | number=1 85 | args=256, 1, 1 86 | 87 | [nn.Upsample] 88 | from=-1 89 | number=1 90 | args='None', 2, 'nearest' 91 | 92 | [Concat] 93 | from=-1, 4 94 | number=1 95 | args=1 96 | 97 | [BottleneckCSP] 98 | from=-1 99 | number=3 100 | args=256, False 101 | 102 | [Conv] 103 | from=-1 104 | number=1 105 | args=256, 3, 2 106 | 107 | [Concat] 108 | from=-1, 14 109 | number=1 110 | args=1 111 | 112 | [BottleneckCSP] 113 | from=-1 114 | number=3 115 | args=512, False 116 | 117 | [Conv] 118 | from=-1 119 | number=1 120 | args=512, 3, 2 121 | 122 | [Concat] 123 | from=-1, 10 124 | number=1 125 | args=1 126 | 127 | [BottleneckCSP] 128 | from=-1 129 | number=3 130 | args=1024, False 131 | 132 | [Detect] 133 | from=17, 20, 23 134 | number=1 135 | args='nc', 'anchors' 136 | -------------------------------------------------------------------------------- /sln/sln.sln: -------------------------------------------------------------------------------- 1 | 2 | Microsoft Visual Studio Solution File, Format Version 12.00 3 | # Visual Studio 14 4 | VisualStudioVersion = 14.0.25420.1 5 | MinimumVisualStudioVersion = 10.0.40219.1 6 | Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "dll_detector", "dll_detector\dll_detector.vcxproj", "{AE246A73-03C7-4A8E-B978-DA5CE9C73B97}" 7 | EndProject 8 | Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "test_dll", "test_dll\test_dll.vcxproj", "{AE9921FC-C594-4C41-925A-5582896BD2A8}" 9 | EndProject 10 | Global 11 | GlobalSection(SolutionConfigurationPlatforms) = preSolution 12 | Debug|x64 = Debug|x64 13 | Debug|x86 = Debug|x86 14 | Release|x64 = Release|x64 15 | Release|x86 = Release|x86 16 | EndGlobalSection 17 | GlobalSection(ProjectConfigurationPlatforms) = postSolution 18 | {AE246A73-03C7-4A8E-B978-DA5CE9C73B97}.Debug|x64.ActiveCfg = Debug|x64 19 | {AE246A73-03C7-4A8E-B978-DA5CE9C73B97}.Debug|x64.Build.0 = Debug|x64 20 | {AE246A73-03C7-4A8E-B978-DA5CE9C73B97}.Debug|x86.ActiveCfg = Debug|Win32 21 | {AE246A73-03C7-4A8E-B978-DA5CE9C73B97}.Debug|x86.Build.0 = Debug|Win32 22 | {AE246A73-03C7-4A8E-B978-DA5CE9C73B97}.Release|x64.ActiveCfg = Release|x64 23 | {AE246A73-03C7-4A8E-B978-DA5CE9C73B97}.Release|x64.Build.0 = Release|x64 24 | {AE246A73-03C7-4A8E-B978-DA5CE9C73B97}.Release|x86.ActiveCfg = Release|Win32 25 | {AE246A73-03C7-4A8E-B978-DA5CE9C73B97}.Release|x86.Build.0 = Release|Win32 26 | {AE9921FC-C594-4C41-925A-5582896BD2A8}.Debug|x64.ActiveCfg = Debug|x64 27 | {AE9921FC-C594-4C41-925A-5582896BD2A8}.Debug|x64.Build.0 = Debug|x64 28 | {AE9921FC-C594-4C41-925A-5582896BD2A8}.Debug|x86.ActiveCfg = Debug|Win32 29 | {AE9921FC-C594-4C41-925A-5582896BD2A8}.Debug|x86.Build.0 = Debug|Win32 30 | {AE9921FC-C594-4C41-925A-5582896BD2A8}.Release|x64.ActiveCfg = Release|x64 31 | {AE9921FC-C594-4C41-925A-5582896BD2A8}.Release|x64.Build.0 = Release|x64 32 | {AE9921FC-C594-4C41-925A-5582896BD2A8}.Release|x86.ActiveCfg = Release|Win32 33 | {AE9921FC-C594-4C41-925A-5582896BD2A8}.Release|x86.Build.0 = Release|Win32 34 | EndGlobalSection 35 | GlobalSection(SolutionProperties) = preSolution 36 | HideSolutionNode = FALSE 37 | EndGlobalSection 38 | EndGlobal 39 | -------------------------------------------------------------------------------- /modules/yolov4.cpp: -------------------------------------------------------------------------------- 1 | 2 | #include "yolov4.h" 3 | 4 | YoloV4::YoloV4( const NetworkInfo &network_info_, 5 | const InferParams &infer_params_) : 6 | Yolo(network_info_, infer_params_) {} 7 | 8 | std::vector YoloV4::decodeTensor(const int imageIdx, const int imageH, const int imageW, const TensorInfo& tensor) 9 | { 10 | float scale_h = 1.f; 11 | float scale_w = 1.f; 12 | int xOffset = 0; 13 | int yOffset = 0; 14 | 15 | const float* detections = &tensor.hostBuffer[imageIdx * tensor.volume]; 16 | 17 | std::vector binfo; 18 | for (uint32_t y = 0; y < tensor.grid_h; ++y) 19 | { 20 | for (uint32_t x = 0; x < tensor.grid_w; ++x) 21 | { 22 | for (uint32_t b = 0; b < tensor.numBBoxes; ++b) 23 | { 24 | const float pw = tensor.anchors[tensor.masks[b] * 2]; 25 | const float ph = tensor.anchors[tensor.masks[b] * 2 + 1]; 26 | 27 | const int numGridCells = tensor.grid_h * tensor.grid_w; 28 | const int bbindex = y * tensor.grid_w + x; 29 | const float bx 30 | = x + detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 0)]; 31 | const float by 32 | = y + detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 1)]; 33 | const float bw 34 | = pw * detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 2)]; 35 | const float bh 36 | = ph * detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 3)]; 37 | 38 | const float objectness 39 | = detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 4)]; 40 | 41 | float maxProb = 0.0f; 42 | int maxIndex = -1; 43 | 44 | for (uint32_t i = 0; i < tensor.numClasses; ++i) 45 | { 46 | float prob 47 | = (detections[bbindex 48 | + numGridCells * (b * (5 + tensor.numClasses) + (5 + i))]); 49 | 50 | if (prob > maxProb) 51 | { 52 | maxProb = prob; 53 | maxIndex = i; 54 | } 55 | } 56 | maxProb = objectness * maxProb; 57 | 58 | if (maxProb > m_ProbThresh) 59 | { 60 | add_bbox_proposal(bx, by, bw, bh, tensor.stride_h, tensor.stride_w, scale_h, scale_w, xOffset, yOffset, maxIndex, maxProb, imageW, imageH, binfo); 61 | } 62 | } 63 | } 64 | } 65 | return binfo; 66 | } 67 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5l6.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.0 # model depth multiple 4 | width_multiple: 1.0 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [ 19,27, 44,40, 38,94 ] # P3/8 9 | - [ 96,68, 86,152, 180,137 ] # P4/16 10 | - [ 140,301, 303,264, 238,542 ] # P5/32 11 | - [ 436,615, 739,380, 925,792 ] # P6/64 12 | 13 | # YOLOv5 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 17 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 18 | [ -1, 3, C3, [ 128 ] ], 19 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 20 | [ -1, 9, C3, [ 256 ] ], 21 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 22 | [ -1, 9, C3, [ 512 ] ], 23 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 24 | [ -1, 3, C3, [ 768 ] ], 25 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 26 | [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ], 27 | [ -1, 3, C3, [ 1024, False ] ], # 11 28 | ] 29 | 30 | # YOLOv5 head 31 | head: 32 | [ [ -1, 1, Conv, [ 768, 1, 1 ] ], 33 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 34 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 35 | [ -1, 3, C3, [ 768, False ] ], # 15 36 | 37 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 38 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 39 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 40 | [ -1, 3, C3, [ 512, False ] ], # 19 41 | 42 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 43 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 44 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 45 | [ -1, 3, C3, [ 256, False ] ], # 23 (P3/8-small) 46 | 47 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 48 | [ [ -1, 20 ], 1, Concat, [ 1 ] ], # cat head P4 49 | [ -1, 3, C3, [ 512, False ] ], # 26 (P4/16-medium) 50 | 51 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 52 | [ [ -1, 16 ], 1, Concat, [ 1 ] ], # cat head P5 53 | [ -1, 3, C3, [ 768, False ] ], # 29 (P5/32-large) 54 | 55 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 56 | [ [ -1, 12 ], 1, Concat, [ 1 ] ], # cat head P6 57 | [ -1, 3, C3, [ 1024, False ] ], # 32 (P6/64-xlarge) 58 | 59 | [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5m6.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.67 # model depth multiple 4 | width_multiple: 0.75 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [ 19,27, 44,40, 38,94 ] # P3/8 9 | - [ 96,68, 86,152, 180,137 ] # P4/16 10 | - [ 140,301, 303,264, 238,542 ] # P5/32 11 | - [ 436,615, 739,380, 925,792 ] # P6/64 12 | 13 | # YOLOv5 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 17 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 18 | [ -1, 3, C3, [ 128 ] ], 19 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 20 | [ -1, 9, C3, [ 256 ] ], 21 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 22 | [ -1, 9, C3, [ 512 ] ], 23 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 24 | [ -1, 3, C3, [ 768 ] ], 25 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 26 | [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ], 27 | [ -1, 3, C3, [ 1024, False ] ], # 11 28 | ] 29 | 30 | # YOLOv5 head 31 | head: 32 | [ [ -1, 1, Conv, [ 768, 1, 1 ] ], 33 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 34 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 35 | [ -1, 3, C3, [ 768, False ] ], # 15 36 | 37 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 38 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 39 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 40 | [ -1, 3, C3, [ 512, False ] ], # 19 41 | 42 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 43 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 44 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 45 | [ -1, 3, C3, [ 256, False ] ], # 23 (P3/8-small) 46 | 47 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 48 | [ [ -1, 20 ], 1, Concat, [ 1 ] ], # cat head P4 49 | [ -1, 3, C3, [ 512, False ] ], # 26 (P4/16-medium) 50 | 51 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 52 | [ [ -1, 16 ], 1, Concat, [ 1 ] ], # cat head P5 53 | [ -1, 3, C3, [ 768, False ] ], # 29 (P5/32-large) 54 | 55 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 56 | [ [ -1, 12 ], 1, Concat, [ 1 ] ], # cat head P6 57 | [ -1, 3, C3, [ 1024, False ] ], # 32 (P6/64-xlarge) 58 | 59 | [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5s6.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 0.33 # model depth multiple 4 | width_multiple: 0.50 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [ 19,27, 44,40, 38,94 ] # P3/8 9 | - [ 96,68, 86,152, 180,137 ] # P4/16 10 | - [ 140,301, 303,264, 238,542 ] # P5/32 11 | - [ 436,615, 739,380, 925,792 ] # P6/64 12 | 13 | # YOLOv5 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 17 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 18 | [ -1, 3, C3, [ 128 ] ], 19 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 20 | [ -1, 9, C3, [ 256 ] ], 21 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 22 | [ -1, 9, C3, [ 512 ] ], 23 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 24 | [ -1, 3, C3, [ 768 ] ], 25 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 26 | [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ], 27 | [ -1, 3, C3, [ 1024, False ] ], # 11 28 | ] 29 | 30 | # YOLOv5 head 31 | head: 32 | [ [ -1, 1, Conv, [ 768, 1, 1 ] ], 33 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 34 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 35 | [ -1, 3, C3, [ 768, False ] ], # 15 36 | 37 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 38 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 39 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 40 | [ -1, 3, C3, [ 512, False ] ], # 19 41 | 42 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 43 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 44 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 45 | [ -1, 3, C3, [ 256, False ] ], # 23 (P3/8-small) 46 | 47 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 48 | [ [ -1, 20 ], 1, Concat, [ 1 ] ], # cat head P4 49 | [ -1, 3, C3, [ 512, False ] ], # 26 (P4/16-medium) 50 | 51 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 52 | [ [ -1, 16 ], 1, Concat, [ 1 ] ], # cat head P5 53 | [ -1, 3, C3, [ 768, False ] ], # 29 (P5/32-large) 54 | 55 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 56 | [ [ -1, 12 ], 1, Concat, [ 1 ] ], # cat head P6 57 | [ -1, 3, C3, [ 1024, False ] ], # 32 (P6/64-xlarge) 58 | 59 | [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /configs/yolov5-5.0/yolov5x6.yaml: -------------------------------------------------------------------------------- 1 | # parameters 2 | nc: 80 # number of classes 3 | depth_multiple: 1.33 # model depth multiple 4 | width_multiple: 1.25 # layer channel multiple 5 | 6 | # anchors 7 | anchors: 8 | - [ 19,27, 44,40, 38,94 ] # P3/8 9 | - [ 96,68, 86,152, 180,137 ] # P4/16 10 | - [ 140,301, 303,264, 238,542 ] # P5/32 11 | - [ 436,615, 739,380, 925,792 ] # P6/64 12 | 13 | # YOLOv5 backbone 14 | backbone: 15 | # [from, number, module, args] 16 | [ [ -1, 1, Focus, [ 64, 3 ] ], # 0-P1/2 17 | [ -1, 1, Conv, [ 128, 3, 2 ] ], # 1-P2/4 18 | [ -1, 3, C3, [ 128 ] ], 19 | [ -1, 1, Conv, [ 256, 3, 2 ] ], # 3-P3/8 20 | [ -1, 9, C3, [ 256 ] ], 21 | [ -1, 1, Conv, [ 512, 3, 2 ] ], # 5-P4/16 22 | [ -1, 9, C3, [ 512 ] ], 23 | [ -1, 1, Conv, [ 768, 3, 2 ] ], # 7-P5/32 24 | [ -1, 3, C3, [ 768 ] ], 25 | [ -1, 1, Conv, [ 1024, 3, 2 ] ], # 9-P6/64 26 | [ -1, 1, SPP, [ 1024, [ 3, 5, 7 ] ] ], 27 | [ -1, 3, C3, [ 1024, False ] ], # 11 28 | ] 29 | 30 | # YOLOv5 head 31 | head: 32 | [ [ -1, 1, Conv, [ 768, 1, 1 ] ], 33 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 34 | [ [ -1, 8 ], 1, Concat, [ 1 ] ], # cat backbone P5 35 | [ -1, 3, C3, [ 768, False ] ], # 15 36 | 37 | [ -1, 1, Conv, [ 512, 1, 1 ] ], 38 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 39 | [ [ -1, 6 ], 1, Concat, [ 1 ] ], # cat backbone P4 40 | [ -1, 3, C3, [ 512, False ] ], # 19 41 | 42 | [ -1, 1, Conv, [ 256, 1, 1 ] ], 43 | [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ], 44 | [ [ -1, 4 ], 1, Concat, [ 1 ] ], # cat backbone P3 45 | [ -1, 3, C3, [ 256, False ] ], # 23 (P3/8-small) 46 | 47 | [ -1, 1, Conv, [ 256, 3, 2 ] ], 48 | [ [ -1, 20 ], 1, Concat, [ 1 ] ], # cat head P4 49 | [ -1, 3, C3, [ 512, False ] ], # 26 (P4/16-medium) 50 | 51 | [ -1, 1, Conv, [ 512, 3, 2 ] ], 52 | [ [ -1, 16 ], 1, Concat, [ 1 ] ], # cat head P5 53 | [ -1, 3, C3, [ 768, False ] ], # 29 (P5/32-large) 54 | 55 | [ -1, 1, Conv, [ 768, 3, 2 ] ], 56 | [ [ -1, 12 ], 1, Concat, [ 1 ] ], # cat head P6 57 | [ -1, 3, C3, [ 1024, False ] ], # 32 (P6/64-xlarge) 58 | 59 | [ [ 23, 26, 29, 32 ], 1, Detect, [ nc, anchors ] ], # Detect(P3, P4, P5, P6) 60 | ] 61 | -------------------------------------------------------------------------------- /samples/sample_detector.cpp: -------------------------------------------------------------------------------- 1 | #include "class_timer.hpp" 2 | #include "class_detector.h" 3 | 4 | #include 5 | #include 6 | 7 | 8 | int main() 9 | { 10 | Config config_v3; 11 | config_v3.net_type = YOLOV3; 12 | config_v3.file_model_cfg = "../configs/yolov3.cfg"; 13 | config_v3.file_model_weights = "../configs/yolov3.weights"; 14 | config_v3.calibration_image_list_file_txt = "../configs/calibration_images.txt"; 15 | config_v3.inference_precison =FP32; 16 | config_v3.detect_thresh = 0.5; 17 | 18 | Config config_v4; 19 | config_v4.net_type = YOLOV4; 20 | config_v4.file_model_cfg = "../configs/yolov4.cfg"; 21 | config_v4.file_model_weights = "../configs/yolov4.weights"; 22 | config_v4.calibration_image_list_file_txt = "../configs/calibration_images.txt"; 23 | config_v4.inference_precison = FP32; 24 | config_v4.detect_thresh = 0.5; 25 | 26 | Config config_v5; 27 | config_v5.net_type = YOLOV5; 28 | config_v5.detect_thresh = 0.5; 29 | config_v5.file_model_cfg = "../configs/yolov5-6.0/yolov5n.cfg"; 30 | config_v5.file_model_weights = "../configs/yolov5-6.0/yolov5n.weights"; 31 | config_v5.calibration_image_list_file_txt = "../configs/calibration_images.txt"; 32 | config_v5.inference_precison = FP32; 33 | 34 | std::unique_ptr detector(new Detector()); 35 | detector->init(config_v5); 36 | cv::Mat image0 = cv::imread("../configs/dog.jpg", cv::IMREAD_UNCHANGED); 37 | cv::Mat image1 = cv::imread("../configs/person.jpg", cv::IMREAD_UNCHANGED); 38 | std::vector batch_res; 39 | Timer timer; 40 | for (;;) 41 | { 42 | //prepare batch data 43 | std::vector batch_img; 44 | cv::Mat temp0 = image0.clone(); 45 | cv::Mat temp1 = image1.clone(); 46 | batch_img.push_back(temp0); 47 | //batch_img.push_back(temp1); 48 | 49 | //detect 50 | timer.reset(); 51 | detector->detect(batch_img, batch_res); 52 | timer.out("detect"); 53 | 54 | //disp 55 | for (int i=0;i m_Bboxes; 58 | 59 | // unaltered original Image 60 | cv::Mat m_OrigImage; 61 | // letterboxed Image given to the network as input 62 | cv::Mat m_LetterboxImage; 63 | // final image marked with the bounding boxes 64 | cv::Mat m_MarkedImage; 65 | }; 66 | 67 | #endif 68 | -------------------------------------------------------------------------------- /modules/calibrator.h: -------------------------------------------------------------------------------- 1 | /** 2 | MIT License 3 | 4 | Copyright (c) 2018 NVIDIA CORPORATION. All rights reserved. 5 | 6 | Permission is hereby granted, free of charge, to any person obtaining a copy 7 | of this software and associated documentation files (the "Software"), to deal 8 | in the Software without restriction, including without limitation the rights 9 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 | copies of the Software, and to permit persons to whom the Software is 11 | furnished to do so, subject to the following conditions: 12 | 13 | The above copyright notice and this permission notice shall be included in all 14 | copies or substantial portions of the Software. 15 | 16 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 | SOFTWARE. 23 | * 24 | */ 25 | #ifndef _CALIBRATOR_H_ 26 | #define _CALIBRATOR_H_ 27 | 28 | #include "NvInfer.h" 29 | #include "ds_image.h" 30 | #include "trt_utils.h" 31 | 32 | class Int8EntropyCalibrator : public nvinfer1::IInt8EntropyCalibrator2 33 | { 34 | public: 35 | Int8EntropyCalibrator(const uint32_t& batchSize, const std::string& calibImages, 36 | const std::string& calibImagesPath, const std::string& calibTableFilePath, 37 | const uint64_t& inputSize, const uint32_t& inputH, const uint32_t& inputW, 38 | const std::string& inputBlobName,const std::string &s_net_type_); 39 | virtual ~Int8EntropyCalibrator(); 40 | 41 | int getBatchSize() const noexcept override { return m_BatchSize; } 42 | bool getBatch(void* bindings[], const char* names[], int nbBindings) noexcept override; 43 | const void* readCalibrationCache(size_t& length) noexcept override; 44 | void writeCalibrationCache(const void* cache, size_t length) noexcept override; 45 | 46 | private: 47 | const uint32_t m_BatchSize; 48 | const uint32_t m_InputH; 49 | const uint32_t m_InputW; 50 | const uint64_t m_InputSize; 51 | const uint64_t m_InputCount; 52 | const std::string m_InputBlobName; 53 | const std::string _s_net_type; 54 | const std::string m_CalibTableFilePath{nullptr}; 55 | uint32_t m_ImageIndex; 56 | bool m_ReadCache{true}; 57 | void* m_DeviceInput{nullptr}; 58 | std::vector m_ImageList; 59 | std::vector m_CalibrationCache; 60 | }; 61 | 62 | #endif -------------------------------------------------------------------------------- /configs/yolov3-tiny.cfg: -------------------------------------------------------------------------------- 1 | [net] 2 | # Testing 3 | batch=1 4 | subdivisions=1 5 | # Training 6 | # batch=64 7 | # subdivisions=2 8 | width=416 9 | height=416 10 | channels=3 11 | momentum=0.9 12 | decay=0.0005 13 | angle=0 14 | saturation = 1.5 15 | exposure = 1.5 16 | hue=.1 17 | 18 | learning_rate=0.001 19 | burn_in=1000 20 | max_batches = 500200 21 | policy=steps 22 | steps=400000,450000 23 | scales=.1,.1 24 | 25 | [convolutional] 26 | batch_normalize=1 27 | filters=16 28 | size=3 29 | stride=1 30 | pad=1 31 | activation=leaky 32 | 33 | [maxpool] 34 | size=2 35 | stride=2 36 | 37 | [convolutional] 38 | batch_normalize=1 39 | filters=32 40 | size=3 41 | stride=1 42 | pad=1 43 | activation=leaky 44 | 45 | [maxpool] 46 | size=2 47 | stride=2 48 | 49 | [convolutional] 50 | batch_normalize=1 51 | filters=64 52 | size=3 53 | stride=1 54 | pad=1 55 | activation=leaky 56 | 57 | [maxpool] 58 | size=2 59 | stride=2 60 | 61 | [convolutional] 62 | batch_normalize=1 63 | filters=128 64 | size=3 65 | stride=1 66 | pad=1 67 | activation=leaky 68 | 69 | [maxpool] 70 | size=2 71 | stride=2 72 | 73 | [convolutional] 74 | batch_normalize=1 75 | filters=256 76 | size=3 77 | stride=1 78 | pad=1 79 | activation=leaky 80 | 81 | [maxpool] 82 | size=2 83 | stride=2 84 | 85 | [convolutional] 86 | batch_normalize=1 87 | filters=512 88 | size=3 89 | stride=1 90 | pad=1 91 | activation=leaky 92 | 93 | [maxpool] 94 | size=2 95 | stride=1 96 | 97 | [convolutional] 98 | batch_normalize=1 99 | filters=1024 100 | size=3 101 | stride=1 102 | pad=1 103 | activation=leaky 104 | 105 | ########### 106 | 107 | [convolutional] 108 | batch_normalize=1 109 | filters=256 110 | size=1 111 | stride=1 112 | pad=1 113 | activation=leaky 114 | 115 | [convolutional] 116 | batch_normalize=1 117 | filters=512 118 | size=3 119 | stride=1 120 | pad=1 121 | activation=leaky 122 | 123 | [convolutional] 124 | size=1 125 | stride=1 126 | pad=1 127 | filters=255 128 | activation=linear 129 | 130 | 131 | 132 | [yolo] 133 | mask = 3,4,5 134 | anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319 135 | classes=80 136 | num=6 137 | jitter=.3 138 | ignore_thresh = .7 139 | truth_thresh = 1 140 | random=1 141 | 142 | [route] 143 | layers = -4 144 | 145 | [convolutional] 146 | batch_normalize=1 147 | filters=128 148 | size=1 149 | stride=1 150 | pad=1 151 | activation=leaky 152 | 153 | [upsample] 154 | stride=2 155 | 156 | [route] 157 | layers = -1, 8 158 | 159 | [convolutional] 160 | batch_normalize=1 161 | filters=256 162 | size=3 163 | stride=1 164 | pad=1 165 | activation=leaky 166 | 167 | [convolutional] 168 | size=1 169 | stride=1 170 | pad=1 171 | filters=255 172 | activation=linear 173 | 174 | [yolo] 175 | mask = 0,1,2 176 | anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319 177 | classes=80 178 | num=6 179 | jitter=.3 180 | ignore_thresh = .7 181 | truth_thresh = 1 182 | random=1 183 | -------------------------------------------------------------------------------- /modules/kernel.cu: -------------------------------------------------------------------------------- 1 | 2 | #include 3 | #include 4 | #include 5 | #include 6 | #include 7 | 8 | inline __device__ float sigmoidGPU(const float& x) { return 1.0f / (1.0f + __expf(-x)); } 9 | 10 | __global__ void gpuYoloLayerV3(const float* input, float* output, const uint32_t grid_h_, 11 | const uint32_t grid_w_, const uint32_t numOutputClasses, 12 | const uint32_t numBBoxes) 13 | { 14 | uint32_t x_id = blockIdx.x * blockDim.x + threadIdx.x; 15 | uint32_t y_id = blockIdx.y * blockDim.y + threadIdx.y; 16 | uint32_t z_id = blockIdx.z * blockDim.z + threadIdx.z; 17 | 18 | if ((x_id >= grid_w_) || (y_id >= grid_h_) || (z_id >= numBBoxes)) 19 | { 20 | return; 21 | } 22 | 23 | const int numGridCells = grid_h_ * grid_w_; 24 | const int bbindex = y_id * grid_w_ + x_id; 25 | 26 | output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 0)] 27 | = sigmoidGPU(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 0)]); 28 | 29 | output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 1)] 30 | = sigmoidGPU(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 1)]); 31 | 32 | output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 2)] 33 | = __expf(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 2)]); 34 | 35 | output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 3)] 36 | = __expf(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 3)]); 37 | 38 | output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 4)] 39 | = sigmoidGPU(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + 4)]); 40 | 41 | for (uint32_t i = 0; i < numOutputClasses; ++i) 42 | { 43 | output[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + (5 + i))] 44 | = sigmoidGPU(input[bbindex + numGridCells * (z_id * (5 + numOutputClasses) + (5 + i))]); 45 | } 46 | } 47 | 48 | cudaError_t cudaYoloLayerV3(const void* input, void* output, const uint32_t& batchSize, 49 | const uint32_t& n_grid_h_,const uint32_t& n_grid_w_, 50 | const uint32_t& numOutputClasses, const uint32_t& numBBoxes, 51 | uint64_t outputSize, cudaStream_t stream) 52 | { 53 | dim3 threads_per_block(16, 16, 4); 54 | dim3 number_of_blocks((n_grid_w_ / threads_per_block.x) + 1, 55 | (n_grid_h_ / threads_per_block.y) + 1, 56 | (numBBoxes / threads_per_block.z) + 1); 57 | for (int batch = 0; batch < batchSize; ++batch) 58 | { 59 | gpuYoloLayerV3<<>>( 60 | reinterpret_cast(input) + (batch * outputSize), 61 | reinterpret_cast(output) + (batch * outputSize), n_grid_h_, n_grid_w_, numOutputClasses, 62 | numBBoxes); 63 | } 64 | return cudaGetLastError(); 65 | } -------------------------------------------------------------------------------- /modules/yoloplugin_lib.h: -------------------------------------------------------------------------------- 1 | /** 2 | MIT License 3 | 4 | Copyright (c) 2018 NVIDIA CORPORATION. All rights reserved. 5 | 6 | Permission is hereby granted, free of charge, to any person obtaining a copy 7 | of this software and associated documentation files (the "Software"), to deal 8 | in the Software without restriction, including without limitation the rights 9 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 | copies of the Software, and to permit persons to whom the Software is 11 | furnished to do so, subject to the following conditions: 12 | 13 | The above copyright notice and this permission notice shall be included in all 14 | copies or substantial portions of the Software. 15 | 16 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 | SOFTWARE. 23 | * 24 | */ 25 | #ifndef __YOLOPLUGIN_LIB__ 26 | #define __YOLOPLUGIN_LIB__ 27 | 28 | //#include 29 | 30 | #include "calibrator.h" 31 | #include "trt_utils.h" 32 | #include "yolo.h" 33 | 34 | //#ifdef __cplusplus 35 | //extern "C" { 36 | //#endif 37 | 38 | #define MAX_OBJECTS_PER_FRAME 32 39 | typedef struct YoloPluginCtx YoloPluginCtx; 40 | typedef struct YoloPluginOutput YoloPluginOutput; 41 | // Init parameters structure as input, required for instantiating yoloplugin_lib 42 | typedef struct 43 | { 44 | // Width at which frame/object will be scaled 45 | int processingWidth; 46 | // height at which frame/object will be scaled 47 | int processingHeight; 48 | // Flag to indicate whether operating on crops of full frame 49 | int fullFrame; 50 | // Plugin config file 51 | std::string configFilePath; 52 | } YoloPluginInitParams; 53 | 54 | struct YoloPluginCtx 55 | { 56 | YoloPluginInitParams initParams; 57 | NetworkInfo networkInfo; 58 | InferParams inferParams; 59 | Yolo* inferenceNetwork; 60 | 61 | // perf vars 62 | float inferTime = 0.0, preTime = 0.0, postTime = 0.0; 63 | uint32_t batchSize = 0; 64 | uint64_t imageCount = 0; 65 | }; 66 | 67 | // Detected/Labelled object structure, stores bounding box info along with label 68 | typedef struct 69 | { 70 | int left; 71 | int top; 72 | int width; 73 | int height; 74 | char label[64]; 75 | } YoloPluginObject; 76 | 77 | // Output data returned after processing 78 | struct YoloPluginOutput 79 | { 80 | int numObjects; 81 | YoloPluginObject object[MAX_OBJECTS_PER_FRAME]; 82 | }; 83 | 84 | // Initialize library context 85 | YoloPluginCtx* YoloPluginCtxInit(YoloPluginInitParams* initParams, size_t batchSize); 86 | 87 | // Dequeue processed output 88 | std::vector YoloPluginProcess(YoloPluginCtx* ctx, std::vector& cvmats); 89 | 90 | // Deinitialize library context 91 | void YoloPluginCtxDeinit(YoloPluginCtx* ctx); 92 | 93 | //#ifdef __cplusplus 94 | //} 95 | //#endif 96 | 97 | #endif 98 | -------------------------------------------------------------------------------- /sln/dll_detector/dll_detector.vcxproj.filters: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | {4FC737F1-C7A5-4376-A066-2A32D752A2FF} 6 | cpp;c;cc;cxx;def;odl;idl;hpj;bat;asm;asmx 7 | 8 | 9 | {93995380-89BD-4b04-88EB-625FBE52EBFB} 10 | h;hh;hpp;hxx;hm;inl;inc;xsd 11 | 12 | 13 | {67DA6AB6-F800-4c08-8B7A-83BB121AAD01} 14 | rc;ico;cur;bmp;dlg;rc2;rct;bin;rgs;gif;jpg;jpeg;jpe;resx;tiff;tif;png;wav;mfcribbon-ms 15 | 16 | 17 | 18 | 19 | Header Files 20 | 21 | 22 | Source Files 23 | 24 | 25 | Source Files 26 | 27 | 28 | 29 | 30 | Source Files 31 | 32 | 33 | Source Files 34 | 35 | 36 | Source Files 37 | 38 | 39 | Source Files 40 | 41 | 42 | Source Files 43 | 44 | 45 | Source Files 46 | 47 | 48 | Source Files 49 | 50 | 51 | Source Files 52 | 53 | 54 | Source Files 55 | 56 | 57 | Source Files 58 | 59 | 60 | Source Files 61 | 62 | 63 | 64 | 65 | Source Files 66 | 67 | 68 | Source Files 69 | 70 | 71 | Source Files 72 | 73 | 74 | Source Files 75 | 76 | 77 | Source Files 78 | 79 | 80 | -------------------------------------------------------------------------------- /modules/chunk.h: -------------------------------------------------------------------------------- 1 | #ifndef CHUNK_H_ 2 | #define CHUNK_H_ 3 | 4 | //#include "NvInfer.h" 5 | //#include "NvInferPlugin.h" 6 | //#include "NvInferRuntimeCommon.h" 7 | //#include 8 | //#include 9 | //#include 10 | //#include 11 | //#include 12 | //#include 13 | //#include 14 | 15 | #include 16 | #include 17 | #include "NvInfer.h" 18 | 19 | namespace nvinfer1 20 | { 21 | class Chunk : public IPluginV2 22 | { 23 | public: 24 | Chunk(); 25 | Chunk(const void* buffer, size_t length); 26 | ~Chunk(); 27 | int getNbOutputs()const noexcept override; 28 | Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims)noexcept override; 29 | int initialize()noexcept override; 30 | void terminate()noexcept override; 31 | size_t getWorkspaceSize(int maxBatchSize) const noexcept override; 32 | // int enqueue(int batchSize, const void* const* inputs, void** outputs, void* workspace, cudaStream_t stream)noexcept override; 33 | 34 | int enqueue(int batchSize, const void* const* inputs, void* const* outputs, void* workspace, 35 | cudaStream_t stream) noexcept override; 36 | 37 | size_t getSerializationSize() const noexcept override; 38 | void serialize(void* buffer) const noexcept override; 39 | const char* getPluginType() const noexcept override; 40 | const char* getPluginVersion() const noexcept override; 41 | void destroy()noexcept override; 42 | void setPluginNamespace(const char* pluginNamespace)noexcept override; 43 | const char* getPluginNamespace() const noexcept override; 44 | DataType getOutputDataType(int index, const nvinfer1::DataType* inputTypes, int nbInputs) const noexcept; 45 | bool isOutputBroadcastAcrossBatch(int outputIndex, const bool* inputIsBroadcasted, int nbInputs) const noexcept; 46 | bool canBroadcastInputAcrossBatch(int inputIndex) const noexcept; 47 | void attachToContext( 48 | cudnnContext* cudnnContext, cublasContext* cublasContext, IGpuAllocator* gpuAllocator); 49 | void configurePlugin(const PluginTensorDesc* in, int nbInput, const PluginTensorDesc* out, int nbOutput); 50 | void detachFromContext(); 51 | bool supportsFormatCombination(int pos, const PluginTensorDesc* inOut, int nbInputs, int nbOutputs) const 52 | { 53 | return inOut[pos].format == TensorFormat::kLINEAR && inOut[pos].type == DataType::kFLOAT; 54 | } 55 | IPluginV2* clone() const noexcept override; 56 | bool supportsFormat(DataType type, PluginFormat format) const noexcept override; 57 | void configureWithFormat(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, DataType type, PluginFormat format, int maxBatchSize) noexcept override; 58 | 59 | private: 60 | std::string _s_plugin_namespace; 61 | int _n_size_split; 62 | }; 63 | 64 | class ChunkPluginCreator : public IPluginCreator 65 | { 66 | public: 67 | ChunkPluginCreator(); 68 | ~ChunkPluginCreator() override = default; 69 | const char* getPluginName()const noexcept override; 70 | const char* getPluginVersion() const noexcept override; 71 | const PluginFieldCollection* getFieldNames()noexcept override; 72 | IPluginV2* createPlugin(const char* name, const PluginFieldCollection* fc)noexcept override; 73 | IPluginV2* deserializePlugin(const char* name, const void* serialData, size_t serialLength)noexcept override; 74 | void setPluginNamespace(const char* libNamespace)noexcept override; 75 | const char* getPluginNamespace() const noexcept override; 76 | private: 77 | std::string _s_name_space; 78 | static PluginFieldCollection _fc; 79 | static std::vector _vec_plugin_attributes; 80 | }; 81 | 82 | }//nampespace nvinfer1 83 | 84 | 85 | #endif 86 | 87 | 88 | -------------------------------------------------------------------------------- /scripts/yaml2cfg.py: -------------------------------------------------------------------------------- 1 | import torch 2 | import argparse 3 | from utils.downloads import * 4 | import struct 5 | import yaml 6 | 7 | def yaml2cfg(opt): 8 | with open(opt.yaml,'r') as f: 9 | with open(opt.yaml.split(".")[0]+".cfg", 'w') as cfg: 10 | f = yaml.load(f,Loader=yaml.FullLoader) 11 | net = 0 12 | for k in f: 13 | if ((k != 'anchors')and(k !='backbone')and(k !='head')): 14 | if net == 0: 15 | cfg.write('\n[net]\n') 16 | cfg.write("width={}\n".format(opt.net_w)) 17 | cfg.write("height={}\n".format(opt.net_h)) 18 | cfg.write("channels={}\n".format(3)) 19 | cfg.write("batch={}\n".format(opt.batch)) 20 | net =1 21 | cfg.write("{}={}\n".format(k,f[k])) 22 | elif (k == 'anchors'): 23 | anchors = [] 24 | for v in f[k]: 25 | anchors.extend(v) 26 | cfg.write("anchors={}\n".format(str(anchors)[1:-1])) 27 | elif((k == 'backbone') or (k =='head')): 28 | for vector in f[k]: 29 | #net type 30 | cfg.write('\n[{}]\n'.format(vector[2])) 31 | #from 32 | if type(vector[0]) == int: 33 | cfg.write("from={}\n".format(str(vector[0]))) 34 | elif type(vector[0]) == list: 35 | cfg.write("from={}\n".format(str(vector[0])[1:-1])) 36 | #number 37 | if type(vector[1]) == int: 38 | cfg.write("number={}\n".format(str(vector[1]))) 39 | elif type(vector[1]) == list: 40 | cfg.write("number={}\n".format(str(vector[1])[1:-1])) 41 | #params 42 | if type(vector[3]) == int: 43 | cfg.write("args={}\n".format(str(vector[3]))) 44 | elif type(vector[3]) == list: 45 | cfg.write("args={}\n".format(str(vector[3])[1:-1])) 46 | 47 | def pt2weight(weights): 48 | model = torch.load(weights, map_location="cpu")['model'].eval().float() 49 | with open(weights.split(".")[0]+".weights",'w') as f: 50 | for k,v in model.state_dict().items(): 51 | list_data = list(v.reshape(-1).numpy()) 52 | f.write('{}'.format(k)) 53 | for value in list_data: 54 | f.write(" ") 55 | f.write(struct.pack('>f',float(value)).hex()) 56 | f.write('\n') 57 | 58 | 59 | if __name__ == '__main__': 60 | parser = argparse.ArgumentParser() 61 | parser.add_argument('--weights', type=str, default='weights/yolov5s.pt', help='model.pt path') 62 | parser.add_argument('--yaml', type=str, default='models/yolov5s.yaml', help='model.yaml path') 63 | parser.add_argument('--net_w', type=int, default=640, help='net width (pixels)') 64 | parser.add_argument('--net_h', type=int, default=320, help='net height (pixels)') 65 | parser.add_argument('--batch', type=int, default=1, help='inference batch size') 66 | parser.add_argument('--classes', type=int,default=80, help='filter by class: --class 0, or --class 0 2 3') 67 | opt = parser.parse_args() 68 | print(opt) 69 | attempt_download(opt.weights) 70 | yaml2cfg(opt) 71 | print("convert {} to {} success".format(opt.yaml,opt.yaml.split(".")[0]+".cfg")) 72 | pt2weight(opt.weights) 73 | print("convert {} to {} success".format(opt.weights,opt.weights.split(".")[0]+".weights")) 74 | print("complete") -------------------------------------------------------------------------------- /modules/yolov3.cpp: -------------------------------------------------------------------------------- 1 | /** 2 | MIT License 3 | 4 | Copyright (c) 2018 NVIDIA CORPORATION. All rights reserved. 5 | 6 | Permission is hereby granted, free of charge, to any person obtaining a copy 7 | of this software and associated documentation files (the "Software"), to deal 8 | in the Software without restriction, including without limitation the rights 9 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 | copies of the Software, and to permit persons to whom the Software is 11 | furnished to do so, subject to the following conditions: 12 | 13 | The above copyright notice and this permission notice shall be included in all 14 | copies or substantial portions of the Software. 15 | 16 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 | SOFTWARE. 23 | * 24 | */ 25 | #include "yolov3.h" 26 | 27 | YoloV3::YoloV3(const NetworkInfo& networkInfo, 28 | const InferParams& inferParams) : 29 | Yolo( networkInfo, inferParams){} 30 | 31 | std::vector YoloV3::decodeTensor(const int imageIdx, 32 | const int imageH, 33 | const int imageW, 34 | const TensorInfo& tensor) 35 | { 36 | float scale_h = 1.f; 37 | float scale_w = 1.f; 38 | int xOffset = 0; 39 | int yOffset = 0; 40 | 41 | const float* detections = &tensor.hostBuffer[imageIdx * tensor.volume]; 42 | 43 | std::vector binfo; 44 | for (uint32_t y = 0; y < tensor.grid_h; ++y) 45 | { 46 | for (uint32_t x = 0; x < tensor.grid_w; ++x) 47 | { 48 | for (uint32_t b = 0; b < tensor.numBBoxes; ++b) 49 | { 50 | const float pw = tensor.anchors[tensor.masks[b] * 2]; 51 | const float ph = tensor.anchors[tensor.masks[b] * 2 + 1]; 52 | 53 | const int numGridCells = tensor.grid_h * tensor.grid_w; 54 | const int bbindex = y * tensor.grid_w+ x; 55 | const float bx 56 | = x + detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 0)]; 57 | 58 | const float by 59 | = y + detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 1)]; 60 | const float bw 61 | = pw * detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 2)]; 62 | const float bh 63 | = ph * detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 3)]; 64 | 65 | const float objectness 66 | = detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 4)]; 67 | 68 | float maxProb = 0.0f; 69 | int maxIndex = -1; 70 | 71 | for (uint32_t i = 0; i < tensor.numClasses; ++i) 72 | { 73 | float prob 74 | = (detections[bbindex 75 | + numGridCells * (b * (5 + tensor.numClasses) + (5 + i))]); 76 | 77 | if (prob > maxProb) 78 | { 79 | maxProb = prob; 80 | maxIndex = i; 81 | } 82 | } 83 | maxProb = objectness * maxProb; 84 | 85 | if (maxProb > m_ProbThresh) 86 | { 87 | add_bbox_proposal(bx, by, bw, bh, tensor.stride_h, tensor.stride_w, scale_h, scale_w, xOffset, yOffset, maxIndex, maxProb, imageW, imageH, binfo); 88 | } 89 | } 90 | } 91 | } 92 | return binfo; 93 | } 94 | -------------------------------------------------------------------------------- /modules/yolov2.cpp: -------------------------------------------------------------------------------- 1 | /** 2 | MIT License 3 | 4 | Copyright (c) 2018 NVIDIA CORPORATION. All rights reserved. 5 | 6 | Permission is hereby granted, free of charge, to any person obtaining a copy 7 | of this software and associated documentation files (the "Software"), to deal 8 | in the Software without restriction, including without limitation the rights 9 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 | copies of the Software, and to permit persons to whom the Software is 11 | furnished to do so, subject to the following conditions: 12 | 13 | The above copyright notice and this permission notice shall be included in all 14 | copies or substantial portions of the Software. 15 | 16 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 | SOFTWARE. 23 | * 24 | */ 25 | 26 | #include "yolov2.h" 27 | #include 28 | 29 | YoloV2::YoloV2(const NetworkInfo& networkInfo, 30 | const InferParams& inferParams) : 31 | Yolo(networkInfo, inferParams){} 32 | 33 | std::vector YoloV2::decodeTensor(const int imageIdx, const int imageH, const int imageW, 34 | const TensorInfo& tensor) 35 | { 36 | float scalingFactor 37 | = std::min(static_cast(m_InputW) / imageW, static_cast(m_InputH) / imageH); 38 | float xOffset = (m_InputW - scalingFactor * imageW) / 2; 39 | float yOffset = (m_InputH - scalingFactor * imageH) / 2; 40 | 41 | float* detections = &tensor.hostBuffer[imageIdx * tensor.volume]; 42 | 43 | std::vector binfo; 44 | for (uint32_t y = 0; y < tensor.gridSize; y++) 45 | { 46 | for (uint32_t x = 0; x < tensor.gridSize; x++) 47 | { 48 | for (uint32_t b = 0; b < tensor.numBBoxes; b++) 49 | { 50 | const float pw = tensor.anchors[2 * b]; 51 | const float ph = tensor.anchors[2 * b + 1]; 52 | const int numGridCells = tensor.gridSize * tensor.gridSize; 53 | const int bbindex = y * tensor.gridSize + x; 54 | const float bx 55 | = x + detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 0)]; 56 | const float by 57 | = y + detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 1)]; 58 | const float bw = pw 59 | * exp(detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 2)]); 60 | const float bh = ph 61 | * exp(detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 3)]); 62 | 63 | const float objectness 64 | = detections[bbindex + numGridCells * (b * (5 + tensor.numClasses) + 4)]; 65 | float maxProb = 0.0f; 66 | int maxIndex = -1; 67 | 68 | for (uint32_t i = 0; i < tensor.numClasses; i++) 69 | { 70 | float prob 71 | = detections[bbindex 72 | + numGridCells * (b * (5 + tensor.numClasses) + (5 + i))]; 73 | 74 | if (prob > maxProb) 75 | { 76 | maxProb = prob; 77 | maxIndex = i; 78 | } 79 | } 80 | 81 | maxProb = objectness * maxProb; 82 | 83 | if (maxProb > m_ProbThresh) 84 | { 85 | addBBoxProposal(bx, by, bw, bh, tensor.stride, scalingFactor, xOffset, yOffset, 86 | maxIndex, maxProb,imageW,imageH, binfo); 87 | } 88 | } 89 | } 90 | } 91 | return binfo; 92 | } 93 | -------------------------------------------------------------------------------- /modules/decodeTensorCUDA.cu: -------------------------------------------------------------------------------- 1 | #include "decodeTensorCUDA.h" 2 | __global__ void decodeTensorKernel( 3 | float* detections, uint32_t* masks, float* anchors, float* boxes, uint32_t grid_h, uint32_t grid_w, uint32_t numClasses, uint32_t numBBoxes) 4 | { 5 | // get idx 6 | uint32_t y = blockIdx.y * blockDim.y + threadIdx.y; 7 | uint32_t x = blockIdx.x * blockDim.x + threadIdx.x; 8 | if (y >= grid_h || x >= grid_w) return; 9 | 10 | const int numGridCells = grid_h * grid_w; 11 | 12 | for (uint32_t b = 0; b < numBBoxes; ++b) 13 | { 14 | const float pw = anchors[masks[b] * 2]; 15 | const float ph = anchors[masks[b] * 2 + 1]; 16 | 17 | // printf("pw %f, ph %f \n", pw, ph); 18 | const uint32_t bbindex = y * grid_w + x; 19 | boxes[18 * bbindex + 6 * b + 0] = x + detections[bbindex + numGridCells * (b * (5 + numClasses) + 0)]; 20 | 21 | boxes[18 * bbindex + 6 * b + 1] = y + detections[bbindex + numGridCells * (b * (5 + numClasses) + 1)]; 22 | boxes[18 * bbindex + 6 * b + 2] = pw * detections[bbindex + numGridCells * (b * (5 + numClasses) + 2)]; 23 | boxes[18 * bbindex + 6 * b + 3] = ph * detections[bbindex + numGridCells * (b * (5 + numClasses) + 3)]; 24 | 25 | // printf("x %f y %f w %f h %f\n", boxes[18 * bbindex + 6 * b + 0], boxes[18 * bbindex + 6 * b + 1], boxes[18 * bbindex + 6 * b + 2], boxes[18 * bbindex + 6 * b + 3]); 26 | 27 | const float objectness = detections[bbindex + numGridCells * (b * (5 + numClasses) + 4)]; 28 | float maxProb = 0.0f; 29 | int maxIndex = -1; 30 | 31 | for (uint32_t i = 0; i < numClasses; ++i) 32 | { 33 | float prob = detections[bbindex + numGridCells * (b * (5 + numClasses) + (5 + i))]; 34 | 35 | if (prob > maxProb) 36 | { 37 | maxProb = prob; 38 | maxIndex = i; 39 | } 40 | } 41 | // printf("objectness * maxProb %f , objectness %f , maxProb %f \n", objectness * maxProb, objectness, maxProb); 42 | boxes[18 * bbindex + 6 * b + 4] = objectness * maxProb; 43 | boxes[18 * bbindex + 6 * b + 5] = (float) maxIndex; 44 | } 45 | } 46 | 47 | float* decodeTensorCUDA(const int imageIdx, const TensorInfo& tensor) 48 | { 49 | // host 50 | int boxes_bytes = 6*sizeof(float)*tensor.grid_h*tensor.grid_w*tensor.numBBoxes; // x y w h maxProb maxIndex 6个元素 51 | const float* detections = &tensor.hostBuffer[imageIdx * tensor.volume]; 52 | float* boxes = (float*) malloc(boxes_bytes); 53 | 54 | uint32_t grid_h = tensor.grid_h; 55 | uint32_t grid_w = tensor.grid_w; 56 | uint32_t numClasses = tensor.numClasses; 57 | uint32_t numBBoxes = tensor.numBBoxes; 58 | 59 | // device 60 | float* d_detections; 61 | int detections_bytes = sizeof(float) * grid_h * grid_w * (5 + numClasses) * numBBoxes; 62 | cudaMalloc((void**) &d_detections, detections_bytes); 63 | cudaMemcpy((void*) d_detections, (void*) detections, detections_bytes, cudaMemcpyHostToDevice); 64 | 65 | uint32_t* d_masks; 66 | cudaMalloc((void**) &d_masks, sizeof(uint32_t)*numBBoxes); 67 | cudaMemcpy((void*) d_masks, (void*) &tensor.masks[0], sizeof(uint32_t)*numBBoxes, cudaMemcpyHostToDevice); 68 | 69 | float* d_anchors; 70 | cudaMalloc((void**) &d_anchors, sizeof(float)*tensor.anchors.size()); 71 | cudaMemcpy((void*) d_anchors, (void*) &tensor.anchors[0], sizeof(float)*tensor.anchors.size(), cudaMemcpyHostToDevice); 72 | 73 | float* d_boxes; 74 | cudaMalloc((void**) &d_boxes, boxes_bytes); 75 | 76 | // CUDA Grid and Block 77 | dim3 threads_per_block(20, 20); 78 | dim3 number_of_blocks((tensor.grid_w / threads_per_block.x) + 1, (tensor.grid_h / threads_per_block.y) + 1); 79 | 80 | // start kernel 81 | decodeTensorKernel<<>>(d_detections, d_masks, d_anchors, d_boxes, grid_h, grid_w, numClasses, numBBoxes); 82 | 83 | // asynchronous copy 84 | cudaMemcpyAsync((void*) boxes, (void*) d_boxes, boxes_bytes, cudaMemcpyDeviceToHost); 85 | 86 | // wait for kernel 87 | cudaDeviceSynchronize(); 88 | 89 | // free memory 90 | cudaFree(d_detections); 91 | cudaFree(d_masks); 92 | cudaFree(d_anchors); 93 | cudaFree(d_boxes); 94 | 95 | // decode results 96 | return boxes; 97 | } 98 | -------------------------------------------------------------------------------- /modules/hardswish.h: -------------------------------------------------------------------------------- 1 | 2 | #ifndef _HARDSWISH_H_ 3 | #define _HARDSWISH_H_ 4 | 5 | #include 6 | #include 7 | #include "NvInfer.h" 8 | 9 | namespace nvinfer1 10 | { 11 | template 12 | void w(char*& buffer, const T& val) 13 | { 14 | *reinterpret_cast(buffer) = val; 15 | buffer += sizeof(T); 16 | } 17 | 18 | template 19 | void r(const char*& buffer, T& val) 20 | { 21 | val = *reinterpret_cast(buffer); 22 | buffer += sizeof(T); 23 | } 24 | 25 | class Hardswish :public IPluginV2 26 | { 27 | public: 28 | Hardswish(); 29 | Hardswish(const void* data, size_t length); 30 | ~Hardswish(); 31 | int getNbOutputs()const noexcept override 32 | { 33 | return 1; 34 | } 35 | Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) noexcept override 36 | { 37 | return inputs[0]; 38 | } 39 | int initialize() noexcept override 40 | { 41 | return 0; 42 | } 43 | void terminate() noexcept override 44 | { 45 | } 46 | size_t getWorkspaceSize(int maxBatchSize) const noexcept override 47 | { 48 | return 0; 49 | } 50 | 51 | bool supportsFormat(DataType type, PluginFormat format) const noexcept override; 52 | void configureWithFormat(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, DataType type, PluginFormat format, int maxBatchSize) noexcept override; 53 | 54 | int enqueue(int batchSize, const void* const* inputs, void* const* outputs, void* workspace, 55 | cudaStream_t stream) noexcept override; 56 | 57 | size_t getSerializationSize() const noexcept override; 58 | void serialize(void* buffer) const noexcept override; 59 | const char* getPluginType() const noexcept override 60 | { 61 | return "HARDSWISH_TRT"; 62 | } 63 | const char* getPluginVersion() const noexcept override 64 | { 65 | return "1.0"; 66 | } 67 | void destroy() noexcept override 68 | { 69 | delete this; 70 | } 71 | void setPluginNamespace(const char* pluginNamespace) noexcept override 72 | { 73 | _s_plugin_namespace = pluginNamespace; 74 | } 75 | const char* getPluginNamespace() const noexcept override 76 | { 77 | return _s_plugin_namespace.c_str(); 78 | } 79 | DataType getOutputDataType(int index, const nvinfer1::DataType* inputTypes, int nbInputs) const noexcept 80 | { 81 | return DataType::kFLOAT; 82 | } 83 | bool isOutputBroadcastAcrossBatch(int outputIndex, const bool* inputIsBroadcasted, int nbInputs) const noexcept 84 | { 85 | return false; 86 | } 87 | bool canBroadcastInputAcrossBatch(int inputIndex) const noexcept 88 | { 89 | return false; 90 | } 91 | void attachToContext( 92 | cudnnContext* cudnnContext, cublasContext* cublasContext, IGpuAllocator* gpuAllocator) noexcept 93 | {} 94 | void configurePlugin(const PluginTensorDesc* in, int nbInput, const PluginTensorDesc* out, int nbOutput) noexcept; 95 | void detachFromContext() noexcept 96 | {} 97 | bool supportsFormatCombination(int pos, const PluginTensorDesc* inOut, int nbInputs, int nbOutputs) const noexcept 98 | { 99 | return inOut[pos].format == TensorFormat::kLINEAR && inOut[pos].type == DataType::kFLOAT; 100 | } 101 | IPluginV2* clone() const noexcept override; 102 | private: 103 | 104 | uint32_t _n_max_thread_pre_block; 105 | uint32_t _n_output_size; 106 | std::string _s_plugin_namespace; 107 | }; //end detect 108 | 109 | class HardswishPluginCreator : public IPluginCreator 110 | { 111 | public: 112 | HardswishPluginCreator(); 113 | ~HardswishPluginCreator() override = default; 114 | const char* getPluginName()const noexcept override; 115 | const char* getPluginVersion() const noexcept override; 116 | const PluginFieldCollection* getFieldNames() noexcept override; 117 | IPluginV2* createPlugin(const char* name, const PluginFieldCollection* fc) noexcept override; 118 | IPluginV2* deserializePlugin(const char* name, const void* serialData, size_t serialLength) noexcept override; 119 | void setPluginNamespace(const char* libNamespace) noexcept override; 120 | const char* getPluginNamespace() const noexcept override; 121 | private: 122 | std::string _s_name_space; 123 | static PluginFieldCollection _fc; 124 | static std::vector _vec_plugin_attributes; 125 | };//end detect creator 126 | 127 | }//end namespace nvinfer1 128 | 129 | 130 | 131 | #endif -------------------------------------------------------------------------------- /modules/mish.h: -------------------------------------------------------------------------------- 1 | #ifndef _MISH_PLUGIN_H 2 | #define _MISH_PLUGIN_H 3 | 4 | #include 5 | #include 6 | #include "NvInfer.h" 7 | 8 | 9 | //https://github.com/wang-xinyu/tensorrtx 10 | namespace nvinfer1 11 | { 12 | class MishPlugin: public IPluginV2 13 | { 14 | public: 15 | explicit MishPlugin(); 16 | MishPlugin(const void* data, size_t length); 17 | 18 | ~MishPlugin(); 19 | 20 | int getNbOutputs() const noexcept override 21 | { 22 | return 1; 23 | } 24 | 25 | Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) noexcept override; 26 | 27 | int initialize() noexcept override; 28 | 29 | virtual void terminate() noexcept override {} 30 | 31 | virtual size_t getWorkspaceSize(int maxBatchSize) const noexcept override { return 0;} 32 | 33 | // virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream); 34 | int enqueue(int batchSize, const void* const* inputs, void* const* outputs, void* workspace, 35 | cudaStream_t stream) noexcept override; 36 | bool supportsFormat(DataType type, PluginFormat format) const noexcept override; 37 | void configureWithFormat(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, DataType type, PluginFormat format, int maxBatchSize) noexcept override; 38 | 39 | virtual size_t getSerializationSize() const noexcept override; 40 | 41 | virtual void serialize(void* buffer) const noexcept override; 42 | 43 | bool supportsFormatCombination(int pos, const PluginTensorDesc* inOut, int nbInputs, int nbOutputs) const noexcept { 44 | return inOut[pos].format == TensorFormat::kLINEAR && inOut[pos].type == DataType::kFLOAT; 45 | } 46 | 47 | const char* getPluginType() const noexcept override; 48 | 49 | const char* getPluginVersion() const noexcept override; 50 | 51 | void destroy() noexcept override; 52 | 53 | IPluginV2* clone() const noexcept override; 54 | 55 | void setPluginNamespace(const char* pluginNamespace) noexcept override; 56 | 57 | const char* getPluginNamespace() const noexcept override; 58 | 59 | DataType getOutputDataType(int index, const nvinfer1::DataType* inputTypes, int nbInputs) const noexcept; 60 | 61 | bool isOutputBroadcastAcrossBatch(int outputIndex, const bool* inputIsBroadcasted, int nbInputs) const noexcept; 62 | 63 | bool canBroadcastInputAcrossBatch(int inputIndex) const noexcept; 64 | 65 | void attachToContext( 66 | cudnnContext* cudnnContext, cublasContext* cublasContext, IGpuAllocator* gpuAllocator)noexcept; 67 | 68 | void configurePlugin(const PluginTensorDesc* in, int nbInput, const PluginTensorDesc* out, int nbOutput)noexcept; 69 | 70 | void detachFromContext()noexcept; 71 | 72 | int input_size_; 73 | private: 74 | void forwardGpu(const float *const * inputs, float* output, cudaStream_t stream, int batchSize = 1); 75 | int thread_count_ = 256; 76 | const char* mPluginNamespace; 77 | }; 78 | 79 | class MishPluginCreator : public IPluginCreator 80 | { 81 | public: 82 | MishPluginCreator(); 83 | 84 | ~MishPluginCreator() override = default; 85 | 86 | const char* getPluginName() const noexcept override; 87 | 88 | const char* getPluginVersion() const noexcept override; 89 | 90 | const PluginFieldCollection* getFieldNames() noexcept override; 91 | 92 | IPluginV2* createPlugin(const char* name, const PluginFieldCollection* fc) noexcept override; 93 | 94 | IPluginV2* deserializePlugin(const char* name, const void* serialData, size_t serialLength) noexcept override; 95 | 96 | void setPluginNamespace(const char* libNamespace) noexcept override; 97 | 98 | const char* getPluginNamespace() const noexcept override; 99 | 100 | private: 101 | std::string mNamespace; 102 | static PluginFieldCollection mFC; 103 | static std::vector mPluginAttributes; 104 | }; 105 | } 106 | #endif 107 | -------------------------------------------------------------------------------- /configs/yolov4-tiny.cfg: -------------------------------------------------------------------------------- 1 | [net] 2 | # Testing 3 | batch=2 4 | subdivisions=1 5 | # Training 6 | #batch=64 7 | #subdivisions=1 8 | width=416 9 | height=320 10 | channels=3 11 | momentum=0.9 12 | decay=0.0005 13 | angle=0 14 | saturation = 1.5 15 | exposure = 1.5 16 | hue=.1 17 | 18 | learning_rate=0.00261 19 | burn_in=1000 20 | max_batches = 500200 21 | policy=steps 22 | steps=400000,450000 23 | scales=.1,.1 24 | 25 | [convolutional] 26 | batch_normalize=1 27 | filters=32 28 | size=3 29 | stride=2 30 | pad=1 31 | activation=leaky 32 | 33 | [convolutional] 34 | batch_normalize=1 35 | filters=64 36 | size=3 37 | stride=2 38 | pad=1 39 | activation=leaky 40 | 41 | [convolutional] 42 | batch_normalize=1 43 | filters=64 44 | size=3 45 | stride=1 46 | pad=1 47 | activation=leaky 48 | 49 | [route] 50 | layers=-1 51 | groups=2 52 | group_id=1 53 | 54 | [convolutional] 55 | batch_normalize=1 56 | filters=32 57 | size=3 58 | stride=1 59 | pad=1 60 | activation=leaky 61 | 62 | [convolutional] 63 | batch_normalize=1 64 | filters=32 65 | size=3 66 | stride=1 67 | pad=1 68 | activation=leaky 69 | 70 | [route] 71 | layers = -1,-2 72 | 73 | [convolutional] 74 | batch_normalize=1 75 | filters=64 76 | size=1 77 | stride=1 78 | pad=1 79 | activation=leaky 80 | 81 | [route] 82 | layers = -6,-1 83 | 84 | [maxpool] 85 | size=2 86 | stride=2 87 | 88 | [convolutional] 89 | batch_normalize=1 90 | filters=128 91 | size=3 92 | stride=1 93 | pad=1 94 | activation=leaky 95 | 96 | [route] 97 | layers=-1 98 | groups=2 99 | group_id=1 100 | 101 | [convolutional] 102 | batch_normalize=1 103 | filters=64 104 | size=3 105 | stride=1 106 | pad=1 107 | activation=leaky 108 | 109 | [convolutional] 110 | batch_normalize=1 111 | filters=64 112 | size=3 113 | stride=1 114 | pad=1 115 | activation=leaky 116 | 117 | [route] 118 | layers = -1,-2 119 | 120 | [convolutional] 121 | batch_normalize=1 122 | filters=128 123 | size=1 124 | stride=1 125 | pad=1 126 | activation=leaky 127 | 128 | [route] 129 | layers = -6,-1 130 | 131 | [maxpool] 132 | size=2 133 | stride=2 134 | 135 | [convolutional] 136 | batch_normalize=1 137 | filters=256 138 | size=3 139 | stride=1 140 | pad=1 141 | activation=leaky 142 | 143 | [route] 144 | layers=-1 145 | groups=2 146 | group_id=1 147 | 148 | [convolutional] 149 | batch_normalize=1 150 | filters=128 151 | size=3 152 | stride=1 153 | pad=1 154 | activation=leaky 155 | 156 | [convolutional] 157 | batch_normalize=1 158 | filters=128 159 | size=3 160 | stride=1 161 | pad=1 162 | activation=leaky 163 | 164 | [route] 165 | layers = -1,-2 166 | 167 | [convolutional] 168 | batch_normalize=1 169 | filters=256 170 | size=1 171 | stride=1 172 | pad=1 173 | activation=leaky 174 | 175 | [route] 176 | layers = -6,-1 177 | 178 | [maxpool] 179 | size=2 180 | stride=2 181 | 182 | [convolutional] 183 | batch_normalize=1 184 | filters=512 185 | size=3 186 | stride=1 187 | pad=1 188 | activation=leaky 189 | 190 | ################################## 191 | 192 | [convolutional] 193 | batch_normalize=1 194 | filters=256 195 | size=1 196 | stride=1 197 | pad=1 198 | activation=leaky 199 | 200 | [convolutional] 201 | batch_normalize=1 202 | filters=512 203 | size=3 204 | stride=1 205 | pad=1 206 | activation=leaky 207 | 208 | [convolutional] 209 | size=1 210 | stride=1 211 | pad=1 212 | filters=255 213 | activation=linear 214 | 215 | 216 | 217 | [yolo] 218 | mask = 3,4,5 219 | anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319 220 | classes=80 221 | num=6 222 | jitter=.3 223 | scale_x_y = 1.05 224 | cls_normalizer=1.0 225 | iou_normalizer=0.07 226 | iou_loss=ciou 227 | ignore_thresh = .7 228 | truth_thresh = 1 229 | random=0 230 | resize=1.5 231 | nms_kind=greedynms 232 | beta_nms=0.6 233 | 234 | [route] 235 | layers = -4 236 | 237 | [convolutional] 238 | batch_normalize=1 239 | filters=128 240 | size=1 241 | stride=1 242 | pad=1 243 | activation=leaky 244 | 245 | [upsample] 246 | stride=2 247 | 248 | [route] 249 | layers = -1, 23 250 | 251 | [convolutional] 252 | batch_normalize=1 253 | filters=256 254 | size=3 255 | stride=1 256 | pad=1 257 | activation=leaky 258 | 259 | [convolutional] 260 | size=1 261 | stride=1 262 | pad=1 263 | filters=255 264 | activation=linear 265 | 266 | [yolo] 267 | mask = 1,2,3 268 | anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319 269 | classes=80 270 | num=6 271 | jitter=.3 272 | scale_x_y = 1.05 273 | cls_normalizer=1.0 274 | iou_normalizer=0.07 275 | iou_loss=ciou 276 | ignore_thresh = .7 277 | truth_thresh = 1 278 | random=0 279 | resize=1.5 280 | nms_kind=greedynms 281 | beta_nms=0.6 282 | -------------------------------------------------------------------------------- /modules/detect.h: -------------------------------------------------------------------------------- 1 | #ifndef _DETECT_H_ 2 | #define _DETECT_H_ 3 | 4 | #include 5 | #include 6 | #include "NvInfer.h" 7 | 8 | namespace nvinfer1 9 | { 10 | template 11 | void write(char*& buffer, const T& val) 12 | { 13 | *reinterpret_cast(buffer) = val; 14 | buffer += sizeof(T); 15 | } 16 | 17 | template 18 | void read(const char*& buffer, T& val) 19 | { 20 | val = *reinterpret_cast(buffer); 21 | buffer += sizeof(T); 22 | } 23 | 24 | class Detect :public IPluginV2 25 | { 26 | public: 27 | Detect(); 28 | Detect(const void* data, size_t length); 29 | Detect(const uint32_t n_anchor_, const uint32_t _n_classes_, 30 | const uint32_t n_grid_h_, const uint32_t n_grid_w_/*, 31 | const uint32_t &n_stride_h_, const uint32_t &n_stride_w_*/); 32 | ~Detect(); 33 | int getNbOutputs()const noexcept override 34 | { 35 | return 1; 36 | } 37 | Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) noexcept override 38 | { 39 | return inputs[0]; 40 | } 41 | int initialize() noexcept override 42 | { 43 | return 0; 44 | } 45 | void terminate() noexcept override 46 | { 47 | } 48 | size_t getWorkspaceSize(int maxBatchSize) const noexcept override 49 | { 50 | return 0; 51 | } 52 | int enqueue(int batchSize, const void* const* inputs, void* const* outputs, void* workspace, 53 | cudaStream_t stream) noexcept override; 54 | 55 | bool supportsFormat(DataType type, PluginFormat format) const noexcept override; 56 | void configureWithFormat(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, DataType type, PluginFormat format, int maxBatchSize) noexcept override; 57 | 58 | size_t getSerializationSize() const noexcept override; 59 | void serialize(void* buffer) const noexcept override; 60 | const char* getPluginType() const noexcept override 61 | { 62 | return "DETECT_TRT"; 63 | } 64 | const char* getPluginVersion() const noexcept override 65 | { 66 | return "1.0"; 67 | } 68 | void destroy() noexcept override 69 | { 70 | delete this; 71 | } 72 | void setPluginNamespace(const char* pluginNamespace) noexcept override 73 | { 74 | _s_plugin_namespace = pluginNamespace; 75 | } 76 | const char* getPluginNamespace() const noexcept override 77 | { 78 | return _s_plugin_namespace.c_str(); 79 | } 80 | DataType getOutputDataType(int index, const nvinfer1::DataType* inputTypes, int nbInputs) const noexcept 81 | { 82 | return DataType::kFLOAT; 83 | } 84 | bool isOutputBroadcastAcrossBatch(int outputIndex, const bool* inputIsBroadcasted, int nbInputs) const noexcept 85 | { 86 | return false; 87 | } 88 | bool canBroadcastInputAcrossBatch(int inputIndex) const noexcept 89 | { 90 | return false; 91 | } 92 | void attachToContext( 93 | cudnnContext* cudnnContext, cublasContext* cublasContext, IGpuAllocator* gpuAllocator) 94 | {} 95 | void configurePlugin(const PluginTensorDesc* in, int nbInput, const PluginTensorDesc* out, int nbOutput) ; 96 | void detachFromContext() 97 | {} 98 | bool supportsFormatCombination(int pos, const PluginTensorDesc* inOut, int nbInputs, int nbOutputs) const noexcept 99 | { 100 | return inOut[pos].format == TensorFormat::kLINEAR && inOut[pos].type == DataType::kFLOAT; 101 | } 102 | IPluginV2* clone() const noexcept override; 103 | private: 104 | 105 | uint32_t _n_anchor; 106 | uint32_t _n_classes; 107 | uint32_t _n_grid_h; 108 | uint32_t _n_grid_w; 109 | //uint32_t _n_stride_h; 110 | // uint32_t _n_stride_w; 111 | uint64_t _n_output_size; 112 | std::string _s_plugin_namespace; 113 | }; //end detect 114 | 115 | class DetectPluginCreator : public IPluginCreator 116 | { 117 | public: 118 | DetectPluginCreator(); 119 | ~DetectPluginCreator() override = default; 120 | const char* getPluginName()const noexcept override; 121 | const char* getPluginVersion() const noexcept override; 122 | const PluginFieldCollection* getFieldNames() noexcept override; 123 | IPluginV2* createPlugin(const char* name, const PluginFieldCollection* fc) noexcept override; 124 | IPluginV2* deserializePlugin(const char* name, const void* serialData, size_t serialLength) noexcept override; 125 | void setPluginNamespace(const char* libNamespace) noexcept override; 126 | const char* getPluginNamespace() const noexcept override; 127 | private: 128 | std::string _s_name_space; 129 | static PluginFieldCollection _fc; 130 | static std::vector _vec_plugin_attributes; 131 | };//end detect creator 132 | 133 | }//end namespace nvinfer1 134 | 135 | 136 | 137 | #endif 138 | -------------------------------------------------------------------------------- /modules/class_yolo_detector.hpp: -------------------------------------------------------------------------------- 1 | #ifndef CLASS_YOLO_DETECTOR_HPP_ 2 | #define CLASS_YOLO_DETECTOR_HPP_ 3 | 4 | #include 5 | #include "ds_image.h" 6 | #include "trt_utils.h" 7 | #include "yolo.h" 8 | #include "yolov2.h" 9 | #include "yolov3.h" 10 | #include "yolov4.h" 11 | #include "yolov5.h" 12 | 13 | #include 14 | #include 15 | #include 16 | #include 17 | #include /* defines FILENAME_MAX */ 18 | 19 | #include "class_detector.h" 20 | #include "class_timer.hpp" 21 | class YoloDectector 22 | { 23 | public: 24 | YoloDectector() 25 | { 26 | 27 | } 28 | ~YoloDectector() 29 | { 30 | 31 | } 32 | 33 | void init(const Config &config) 34 | { 35 | _config = config; 36 | 37 | this->set_gpu_id(_config.gpu_id); 38 | 39 | this->parse_config(); 40 | 41 | this->build_net(); 42 | } 43 | 44 | void detect(const std::vector &vec_image, 45 | std::vector &vec_batch_result) 46 | { 47 | Timer timer; 48 | std::vector vec_ds_images; 49 | vec_batch_result.clear(); 50 | vec_batch_result.resize(vec_image.size()); 51 | for (const auto &img:vec_image) 52 | { 53 | vec_ds_images.emplace_back(img, _vec_net_type[_config.net_type], _p_net->getInputH(), _p_net->getInputW()); 54 | } 55 | cv::Mat trtInput = blobFromDsImages(vec_ds_images, _p_net->getInputH(),_p_net->getInputW()); 56 | timer.out("pre"); 57 | _p_net->doInference(trtInput.data, vec_ds_images.size()); 58 | timer.reset(); 59 | for (uint32_t i = 0; i < vec_ds_images.size(); ++i) 60 | { 61 | auto curImage = vec_ds_images.at(i); 62 | auto binfo = _p_net->decodeDetections(i, curImage.getImageHeight(), curImage.getImageWidth()); 63 | auto remaining = nmsAllClasses(_p_net->getNMSThresh(), 64 | binfo, 65 | _p_net->getNumClasses(), 66 | _vec_net_type[_config.net_type]); 67 | if (remaining.empty()) 68 | { 69 | continue; 70 | } 71 | std::vector vec_result(0); 72 | for (const auto &b : remaining) 73 | { 74 | Result res; 75 | res.id = b.label; 76 | res.prob = b.prob; 77 | const int x = b.box.x1; 78 | const int y = b.box.y1; 79 | const int w = b.box.x2 - b.box.x1; 80 | const int h = b.box.y2 - b.box.y1; 81 | res.rect = cv::Rect(x, y, w, h); 82 | vec_result.push_back(res); 83 | } 84 | vec_batch_result[i] = vec_result; 85 | } 86 | timer.out("post"); 87 | } 88 | 89 | private: 90 | 91 | void set_gpu_id(const int id = 0) 92 | { 93 | cudaError_t status = cudaSetDevice(id); 94 | if (status != cudaSuccess) 95 | { 96 | std::cout << "gpu id :" + std::to_string(id) + " not exist !" << std::endl; 97 | assert(0); 98 | } 99 | } 100 | 101 | void parse_config() 102 | { 103 | _yolo_info.networkType = _vec_net_type[_config.net_type]; 104 | _yolo_info.configFilePath = _config.file_model_cfg; 105 | _yolo_info.wtsFilePath = _config.file_model_weights; 106 | _yolo_info.precision = _vec_precision[_config.inference_precison]; 107 | _yolo_info.deviceType = "kGPU"; 108 | auto npos = _yolo_info.wtsFilePath.find(".weights"); 109 | assert(npos != std::string::npos 110 | && "wts file file not recognised. File needs to be of '.weights' format"); 111 | _yolo_info.data_path = _yolo_info.wtsFilePath.substr(0, npos); 112 | _yolo_info.calibrationTablePath = _yolo_info.data_path + "-calibration.table"; 113 | _yolo_info.inputBlobName = "data"; 114 | 115 | _infer_param.printPerfInfo = false; 116 | _infer_param.printPredictionInfo = false; 117 | _infer_param.calibImages = _config.calibration_image_list_file_txt; 118 | _infer_param.calibImagesPath = ""; 119 | _infer_param.probThresh = _config.detect_thresh; 120 | _infer_param.nmsThresh = 0.5; 121 | } 122 | 123 | void build_net() 124 | { 125 | if (_config.net_type == YOLOV3) 126 | { 127 | _p_net = std::unique_ptr{ new YoloV3(_yolo_info, _infer_param) }; 128 | } 129 | else if( _config.net_type == YOLOV4) 130 | { 131 | _p_net = std::unique_ptr{ new YoloV4(_yolo_info,_infer_param) }; 132 | } 133 | else if (_config.net_type == YOLOV5) 134 | { 135 | _p_net = std::unique_ptr{ new YoloV5(_yolo_info,_infer_param) }; 136 | } 137 | else 138 | { 139 | assert(false && "Unrecognised network_type."); 140 | } 141 | } 142 | 143 | private: 144 | Config _config; 145 | NetworkInfo _yolo_info; 146 | InferParams _infer_param; 147 | std::vector _vec_net_type{ "yolov3","yolov4","yolov5" }; 148 | std::vector _vec_precision{ "kINT8","kHALF","kFLOAT" }; 149 | std::unique_ptr _p_net = nullptr; 150 | Timer _m_timer; 151 | }; 152 | 153 | 154 | #endif 155 | -------------------------------------------------------------------------------- /modules/calibrator.cpp: -------------------------------------------------------------------------------- 1 | /** 2 | MIT License 3 | 4 | Copyright (c) 2018 NVIDIA CORPORATION. All rights reserved. 5 | 6 | Permission is hereby granted, free of charge, to any person obtaining a copy 7 | of this software and associated documentation files (the "Software"), to deal 8 | in the Software without restriction, including without limitation the rights 9 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 | copies of the Software, and to permit persons to whom the Software is 11 | furnished to do so, subject to the following conditions: 12 | 13 | The above copyright notice and this permission notice shall be included in all 14 | copies or substantial portions of the Software. 15 | 16 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 22 | SOFTWARE. 23 | * 24 | */ 25 | 26 | #include "calibrator.h" 27 | #include 28 | #include 29 | #include 30 | #include 31 | #include 32 | Int8EntropyCalibrator::Int8EntropyCalibrator(const uint32_t& batchSize, const std::string& calibImages, 33 | const std::string& calibImagesPath, 34 | const std::string& calibTableFilePath, 35 | const uint64_t& inputSize, const uint32_t& inputH, 36 | const uint32_t& inputW, const std::string& inputBlobName, 37 | const std::string &s_net_type_) : 38 | m_BatchSize(batchSize), 39 | m_InputH(inputH), 40 | m_InputW(inputW), 41 | m_InputSize(inputSize), 42 | m_InputCount(batchSize * inputSize), 43 | m_InputBlobName(inputBlobName), 44 | m_CalibTableFilePath(calibTableFilePath), 45 | m_ImageIndex(0), 46 | _s_net_type(s_net_type_) 47 | { 48 | if (!fileExists(m_CalibTableFilePath, false)) 49 | { 50 | m_ImageList = loadImageList(calibImages, calibImagesPath); 51 | m_ImageList.resize(static_cast(m_ImageList.size() / m_BatchSize) * m_BatchSize); 52 | std::random_shuffle(m_ImageList.begin(), m_ImageList.end(), 53 | [](int i) { return rand() % i; }); 54 | } 55 | 56 | NV_CUDA_CHECK(cudaMalloc(&m_DeviceInput, m_InputCount * sizeof(float))); 57 | } 58 | 59 | Int8EntropyCalibrator::~Int8EntropyCalibrator() { NV_CUDA_CHECK(cudaFree(m_DeviceInput)); } 60 | 61 | bool Int8EntropyCalibrator::getBatch(void* bindings[], const char* names[], int nbBindings)noexcept 62 | { 63 | if (m_ImageIndex + m_BatchSize >= m_ImageList.size()) return false; 64 | 65 | // Load next batch 66 | std::vector dsImages(m_BatchSize); 67 | for (uint32_t j = m_ImageIndex; j < m_ImageIndex + m_BatchSize; ++j) 68 | { 69 | dsImages.at(j - m_ImageIndex) = DsImage(m_ImageList.at(j), _s_net_type, m_InputH, m_InputW); 70 | } 71 | m_ImageIndex += m_BatchSize; 72 | 73 | cv::Mat trtInput = blobFromDsImages(dsImages, m_InputH, m_InputW); 74 | 75 | NV_CUDA_CHECK(cudaMemcpy(m_DeviceInput, trtInput.ptr(0), m_InputCount * sizeof(float), 76 | cudaMemcpyHostToDevice)); 77 | assert(!strcmp(names[0], m_InputBlobName.c_str())); 78 | bindings[0] = m_DeviceInput; 79 | return true; 80 | } 81 | 82 | const void* Int8EntropyCalibrator::readCalibrationCache(size_t& length) noexcept 83 | { 84 | void* output; 85 | m_CalibrationCache.clear(); 86 | assert(!m_CalibTableFilePath.empty()); 87 | std::ifstream input(m_CalibTableFilePath, std::ios::binary | std::ios::in); 88 | input >> std::noskipws; 89 | if (m_ReadCache && input.good()) 90 | std::copy(std::istream_iterator(input), std::istream_iterator(), 91 | std::back_inserter(m_CalibrationCache)); 92 | 93 | length = m_CalibrationCache.size(); 94 | if (length) 95 | { 96 | std::cout << "Using cached calibration table to build the engine" << std::endl; 97 | output = &m_CalibrationCache[0]; 98 | } 99 | 100 | else 101 | { 102 | std::cout << "New calibration table will be created to build the engine" << std::endl; 103 | output = nullptr; 104 | } 105 | 106 | return output; 107 | } 108 | 109 | void Int8EntropyCalibrator::writeCalibrationCache(const void* cache, size_t length) noexcept 110 | { 111 | assert(!m_CalibTableFilePath.empty()); 112 | std::ofstream output(m_CalibTableFilePath, std::ios::binary); 113 | output.write(reinterpret_cast(cache), length); 114 | output.close(); 115 | } -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Yolov5 Yolov4 Yolov3 TensorRT Implementation 2 | 3 | ![GitHub stars](https://img.shields.io/github/stars/enazoe/yolo-tensorrt) ![GitHub forks](https://img.shields.io/github/forks/enazoe/yolo-tensorrt) ![GitHub watchers](https://img.shields.io/github/watchers/enazoe/yolo-tensorrt) [![Gitter](https://badges.gitter.im/yolo-tensorrt/community.svg)](https://gitter.im/yolo-tensorrt/community?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge) 4 | 5 | __news:__ 2021.10.31:yolov5-v6.0 support 6 | 7 | ![](./configs/result.jpg) 8 | ## INTRODUCTION 9 | 10 | The project is the encapsulation of nvidia official yolo-tensorrt [implementation](https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps). And you must have the trained yolo model(__.weights__) and __.cfg__ file from the darknet (yolov3 & yolov4). For the [yolov5](https://github.com/ultralytics/yolov5) ,you should prepare the model file (yolov5s.yaml) and the trained weight file (yolov5s.pt) from pytorch. 11 | 12 | ![](./configs/yolo-trt.png) 13 | 14 | - [x] yolov5n ,yolov5s , yolov5m , yolov5l , yolov5x ,yolov5-p6 [tutorial](yolov5_tutorial.md) 15 | - [x] yolov4 16 | - [x] yolov3 17 | 18 | ## Features 19 | 20 | - [x] inequal net width and height 21 | - [x] batch inference 22 | - [x] support FP32,FP16,INT8 23 | - [ ] dynamic input size 24 | 25 | 26 | ## PLATFORM & BENCHMARK 27 | 28 | - [x] windows 10 29 | - [x] ubuntu 18.04 30 | - [x] L4T (Jetson platform) 31 | 32 |
BENCHMARK 33 | 34 | #### x86 (inference time) 35 | 36 | 37 | | model | size | gpu | fp32 | fp16 | INT8 | 38 | | :-----: | :-----: | :----: | :--: | :--: | :--: | 39 | | yolov5s | 640x640 | 1080ti | 8ms | / | 7ms | 40 | | yolov5m | 640x640 | 1080ti | 13ms | / | 11ms | 41 | | yolov5l | 640x640 | 1080ti | 20ms | / | 15ms | 42 | | yolov5x | 640x640 | 1080ti | 30ms | / | 23ms | 43 | #### Jetson NX with Jetpack4.4.1 (inference / detect time) 44 | 45 | | model | size | gpu | fp32 | fp16 | INT8 | 46 | | :-------------: | :----: | :--: | :--: | :--: | :--: | 47 | | yolov3 | 416x416 | nx | 105ms/120ms | 30ms/48ms | 20ms/35ms | 48 | | yolov3-tiny | 416x416 | nx | 14ms/23ms | 8ms/15ms | 12ms/19ms | 49 | | yolov4-tiny | 416x416 | nx | 13ms/23ms | 7ms/16ms | 7ms/15ms | 50 | | yolov4 | 416x416 | nx | 111ms/125ms | 55ms/65ms | 47ms/57ms | 51 | | yolov5s | 416x416 | nx | 47ms/88ms | 33ms/74ms | 28ms/64ms | 52 | | yolov5m | 416x416 | nx | 110ms/145ms | 63ms/101ms | 49ms/91ms | 53 | | yolov5l | 416x416 | nx | 205ms/242ms | 95ms/123ms | 76ms/118ms | 54 | | yolov5x | 416x416 | nx | 351ms/405ms | 151ms/183ms | 114ms/149ms | 55 | 56 | 57 | ### ubuntu 58 | | model | size | gpu | fp32 | fp16 | INT8 | 59 | | :-------------: | :----: | :--: | :--: | :--: | :--: | 60 | | yolov4 | 416x416 | titanv | 11ms/17ms | 8ms/15ms | 7ms/14ms | 61 | | yolov5s | 416x416 | titanv | 7ms/22ms | 5ms/20ms | 5ms/18ms | 62 | | yolov5m | 416x416 | titanv | 9ms/23ms | 8ms/22ms | 7ms/21ms | 63 | | yolov5l | 416x416 | titanv | 17ms/28ms | 11ms/23ms | 11ms/24ms | 64 | | yolov5x | 416x416 | titanv | 25ms/40ms | 15ms/27ms | 15ms/27ms | 65 |
66 | 67 | ## WRAPPER 68 | 69 | Prepare the pretrained __.weights__ and __.cfg__ model. 70 | 71 | ```c++ 72 | Detector detector; 73 | Config config; 74 | 75 | std::vector res; 76 | detector.detect(vec_image, res) 77 | ``` 78 | 79 | ## Build and use yolo-trt as DLL or SO libraries 80 | 81 | 82 | ### windows10 83 | 84 | - dependency : TensorRT 7.1.3.4 , cuda 11.0 , cudnn 8.0 , opencv4 , vs2015 85 | - build: 86 | 87 | open MSVC _sln/sln.sln_ file 88 | - dll project : the trt yolo detector dll 89 | - demo project : test of the dll 90 | 91 | ### ubuntu & L4T (jetson) 92 | 93 | The project generate the __libdetector.so__ lib, and the sample code. 94 | **_If you want to use the libdetector.so lib in your own project,this [cmake file](https://github.com/enazoe/yolo-tensorrt/blob/master/scripts/CMakeLists.txt) perhaps could help you ._** 95 | 96 | 97 | ```bash 98 | git clone https://github.com/enazoe/yolo-tensorrt.git 99 | cd yolo-tensorrt/ 100 | mkdir build 101 | cd build/ 102 | cmake .. 103 | make 104 | ./yolo-trt 105 | ``` 106 | ## API 107 | 108 | ```c++ 109 | struct Config 110 | { 111 | std::string file_model_cfg = "configs/yolov4.cfg"; 112 | 113 | std::string file_model_weights = "configs/yolov4.weights"; 114 | 115 | float detect_thresh = 0.9; 116 | 117 | ModelType net_type = YOLOV4; 118 | 119 | Precision inference_precison = INT8; 120 | 121 | int gpu_id = 0; 122 | 123 | std::string calibration_image_list_file_txt = "configs/calibration_images.txt"; 124 | 125 | }; 126 | 127 | class API Detector 128 | { 129 | public: 130 | explicit Detector(); 131 | ~Detector(); 132 | 133 | void init(const Config &config); 134 | 135 | void detect(const std::vector &mat_image,std::vector &vec_batch_result); 136 | 137 | private: 138 | Detector(const Detector &); 139 | const Detector &operator =(const Detector &); 140 | class Impl; 141 | Impl *_impl; 142 | }; 143 | ``` 144 | 145 | ## REFERENCE 146 | 147 | - https://github.com/wang-xinyu/tensorrtx/tree/master/yolov4 148 | - https://github.com/mj8ac/trt-yolo-app_win64 149 | - https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps 150 | 151 | ## Contact 152 | 153 | 微信关注公众号EigenVison，回复yolo获取交流群号 154 | 155 | 156 | 157 | ![](./configs/qrcode.jpeg) 158 | -------------------------------------------------------------------------------- /modules/hardswish.cu: -------------------------------------------------------------------------------- 1 | //sys 2 | #include 3 | #include 4 | #include 5 | #include 6 | #include 7 | #include 8 | #include 9 | #include 10 | //my 11 | #include "hardswish.h" 12 | 13 | #define NV_CUDA_CHECK(status) \ 14 | { \ 15 | if (status != 0) \ 16 | { \ 17 | std::cout << "Cuda failure: " << cudaGetErrorString(status) << " in file " << __FILE__ \ 18 | << " at line " << __LINE__ << std::endl; \ 19 | abort(); \ 20 | } \ 21 | } 22 | 23 | namespace nvinfer1 24 | { 25 | Hardswish::Hardswish() 26 | { 27 | cudaDeviceProp prop; 28 | cudaGetDeviceProperties(&prop, 0); 29 | _n_max_thread_pre_block = prop.maxThreadsPerBlock; 30 | // printf("Hardswish():%d\n", _n_max_thread_pre_block); 31 | } 32 | 33 | Hardswish::Hardswish(const void* data, size_t length) 34 | { 35 | const char *d = reinterpret_cast(data), *a = d; 36 | r(d, _n_max_thread_pre_block); 37 | r(d, _n_output_size); 38 | // printf("r:threads:%d,size:%d\n", _n_max_thread_pre_block, _n_output_size); 39 | assert(d == a + length); 40 | } 41 | 42 | Hardswish::~Hardswish() 43 | {} 44 | 45 | __global__ void kernel_hardswish(const float *input_, float *output_, int n_data_size_) 46 | { 47 | int i = threadIdx.x + blockIdx.x * blockDim.x; 48 | if (i >= n_data_size_)return; 49 | if (input_[i] >= 3.0f) 50 | { 51 | output_[i] = input_[i]; 52 | } 53 | else if (input_[i] <= -3.0f) 54 | { 55 | output_[i] = 0.0f; 56 | } 57 | else 58 | { 59 | output_[i] = input_[i] * (input_[i] + 3.0f) / 6.0f; 60 | } 61 | } 62 | 63 | cudaError_t cuda_hardswish_layer(const void* input_, 64 | void* output_, 65 | const int n_batch_size_, 66 | const int n_output_size_, 67 | const int threads_, 68 | cudaStream_t stream_) 69 | { 70 | int n_data_size = n_batch_size_ * n_output_size_; 71 | // printf("cuda_hardswish_layer:%d,size:%d\n", n_batch_size_, n_output_size_); 72 | kernel_hardswish << <(n_data_size + threads_ -1)/threads_, threads_, 0, stream_ >> >( 73 | reinterpret_cast(input_), 74 | reinterpret_cast(output_), 75 | n_data_size); 76 | return cudaGetLastError(); 77 | } 78 | 79 | int Hardswish::enqueue(int batchSize, 80 | const void* const* inputs, 81 | void* const* outputs, 82 | void* workspace, 83 | cudaStream_t stream) noexcept 84 | { 85 | // printf("batch_size:%d,output_size:%d,threads:%d\n", batchSize, _n_output_size, _n_max_thread_pre_block); 86 | NV_CUDA_CHECK(cuda_hardswish_layer(inputs[0], outputs[0], batchSize, _n_output_size , _n_max_thread_pre_block, stream)); 87 | return 0; 88 | } 89 | 90 | size_t Hardswish::getSerializationSize() const noexcept 91 | { 92 | return sizeof(_n_max_thread_pre_block) +sizeof(_n_output_size); 93 | } 94 | 95 | void Hardswish::serialize(void *buffer) const noexcept 96 | { 97 | char *d = static_cast(buffer), *a = d; 98 | w(d, _n_max_thread_pre_block); 99 | w(d, _n_output_size); 100 | // printf("serialize:%d,%d\n", _n_max_thread_pre_block, _n_output_size); 101 | assert(d == a + getSerializationSize()); 102 | } 103 | 104 | 105 | bool Hardswish::supportsFormat(DataType type, PluginFormat format) const noexcept 106 | { 107 | return (type == DataType::kFLOAT && format == PluginFormat::kLINEAR); 108 | } 109 | 110 | void Hardswish::configureWithFormat(const Dims* inputDims, int nbInputs, 111 | const Dims* outputDims, int nbOutputs, DataType type, PluginFormat format, int maxBatchSize) noexcept 112 | { 113 | 114 | } 115 | 116 | 117 | void Hardswish::configurePlugin(const PluginTensorDesc* in, int nbInput, const PluginTensorDesc* out, int nbOutput) noexcept 118 | { 119 | 120 | _n_output_size = in->dims.d[0] * in->dims.d[1] * in->dims.d[2]; 121 | // printf("configurePlugin:%d,%d,%d\n", in->dims.d[0], in->dims.d[1], in->dims.d[2]); 122 | } 123 | IPluginV2* Hardswish::clone() const noexcept 124 | { 125 | Hardswish *p = new Hardswish(); 126 | p->setPluginNamespace(_s_plugin_namespace.c_str()); 127 | p->_n_max_thread_pre_block = _n_max_thread_pre_block; 128 | p->_n_output_size = _n_output_size; 129 | // printf("clone:%d,%d\n", _n_max_thread_pre_block, _n_output_size); 130 | return p; 131 | } 132 | 133 | 134 | // 135 | PluginFieldCollection HardswishPluginCreator::_fc{}; 136 | std::vector HardswishPluginCreator::_vec_plugin_attributes; 137 | 138 | HardswishPluginCreator::HardswishPluginCreator() 139 | { 140 | _vec_plugin_attributes.clear(); 141 | _fc.nbFields = _vec_plugin_attributes.size(); 142 | _fc.fields = _vec_plugin_attributes.data(); 143 | } 144 | 145 | const char* HardswishPluginCreator::getPluginName() const noexcept 146 | { 147 | return "HARDSWISH_TRT"; 148 | } 149 | 150 | const char* HardswishPluginCreator::getPluginVersion() const noexcept 151 | { 152 | return "1.0"; 153 | } 154 | 155 | const PluginFieldCollection* HardswishPluginCreator::getFieldNames() noexcept 156 | { 157 | return &_fc; 158 | } 159 | 160 | IPluginV2* HardswishPluginCreator::createPlugin(const char* name, const PluginFieldCollection* fc) noexcept 161 | { 162 | Hardswish* obj = new Hardswish(); 163 | obj->setPluginNamespace(_s_name_space.c_str()); 164 | return obj; 165 | } 166 | 167 | IPluginV2* HardswishPluginCreator::deserializePlugin(const char* name, const void* serialData, size_t serialLength) noexcept 168 | { 169 | Hardswish* obj = new Hardswish(serialData, serialLength); 170 | obj->setPluginNamespace(_s_name_space.c_str()); 171 | return obj; 172 | } 173 | 174 | void HardswishPluginCreator::setPluginNamespace(const char* libNamespace) noexcept 175 | { 176 | _s_name_space = libNamespace; 177 | } 178 | 179 | const char* HardswishPluginCreator::getPluginNamespace() const noexcept 180 | { 181 | return _s_name_space.c_str(); 182 | } 183 | }//end namespace nvinfer1 184 | -------------------------------------------------------------------------------- /sln/test_dll/test_dll.vcxproj: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | Debug 6 | Win32 7 | 8 | 9 | Release 10 | Win32 11 | 12 | 13 | Debug 14 | x64 15 | 16 | 17 | Release 18 | x64 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | {AE9921FC-C594-4C41-925A-5582896BD2A8} 29 | test_dll 30 | 10.0.17763.0 31 | 32 | 33 | 34 | Application 35 | true 36 | v141 37 | MultiByte 38 | 39 | 40 | Application 41 | false 42 | v141 43 | true 44 | MultiByte 45 | 46 | 47 | Application 48 | true 49 | v141 50 | MultiByte 51 | 52 | 53 | Application 54 | false 55 | v141 56 | true 57 | MultiByte 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | Level3 84 | Disabled 85 | true 86 | 87 | 88 | 89 | 90 | Level3 91 | Disabled 92 | true 93 | 94 | 95 | 96 | 97 | Level3 98 | MaxSpeed 99 | true 100 | true 101 | true 102 | 103 | 104 | true 105 | true 106 | 107 | 108 | 109 | 110 | Level3 111 | MaxSpeed 112 | true 113 | true 114 | true 115 | ..\..\extra;%(AdditionalIncludeDirectories) 116 | 117 | 118 | true 119 | true 120 | detector.lib;%(AdditionalDependencies) 121 | 122 | 123 | 124 | 125 | 126 | 127 | --------------------------------------------------------------------------------