initialization

README.md

@@ -0,0 +1,18 @@
+### 我爱996
+
+#### 我们的工作共分为三部分：
+
+##### 一、训练FasterRCNN模型
+
+>详情请阅读tf-faster-rcnn中的README.md
+
+##### 二、tf模型转换Onnx模型
+
+>详情请阅读tf-faster-rcnn中的faster_rcnn模型转化.md
+
+##### 三、TensorRT加速
+
+>详情请阅读TensorRT-Base中的README.md
+
+
+

TensorRT-Base/.gitignore

@@ -0,0 +1,8 @@
+##ignore this file##
+bin/Release/*
+build/*
+lib/Release/*
+lib/*.so
+mnist_onnx/model/model.onnx
+faster_rcnn/model/VGG16_faster_rcnn_final.caffemodel
+faster_rcnn/model/faster_rcnn.bin

TensorRT-Base/.vscode/c_cpp_properties.json

@@ -0,0 +1,26 @@
+{
+    "configurations": [
+        {
+            "name": "Win32",
+            "includePath": [
+                "${workspaceFolder}/**"
+                // "/usr/local/opencv/include",
+                // "/usr/local/opencv/include/opencv",
+                // "/usr/local/opencv/include/opencv2",
+                // "/usr/local/cuda/include",
+                // "/opt/TensorRT-7.2.2.3/include"
+            ],
+            "defines": [
+                "_DEBUG",
+                "UNICODE",
+                "_UNICODE"
+            ],
+            "compilerPath": "/usr/bin/gcc",
+            "cStandard": "c11",
+            "cppStandard": "c++17",
+            "intelliSenseMode": "clang-x64",
+            "configurationProvider": "ms-vscode.cmake-tools"
+        }
+    ],
+    "version": 4
+}

TensorRT-Base/.vscode/settings.json

@@ -0,0 +1,87 @@
+{
+    "files.associations": {
+        "iosfwd": "cpp",
+        "iostream": "cpp",
+        "*.tcc": "cpp",
+        "chrono": "cpp",
+        "functional": "cpp",
+        "iterator": "cpp",
+        "map": "cpp",
+        "numeric": "cpp",
+        "ratio": "cpp",
+        "cstring": "cpp",
+        "xstring": "cpp",
+        "cmath": "cpp",
+        "array": "cpp",
+        "atomic": "cpp",
+        "cctype": "cpp",
+        "clocale": "cpp",
+        "cstdarg": "cpp",
+        "cstddef": "cpp",
+        "cstdint": "cpp",
+        "cstdio": "cpp",
+        "cstdlib": "cpp",
+        "ctime": "cpp",
+        "cwchar": "cpp",
+        "cwctype": "cpp",
+        "deque": "cpp",
+        "list": "cpp",
+        "unordered_map": "cpp",
+        "vector": "cpp",
+        "exception": "cpp",
+        "algorithm": "cpp",
+        "memory": "cpp",
+        "memory_resource": "cpp",
+        "optional": "cpp",
+        "string": "cpp",
+        "string_view": "cpp",
+        "system_error": "cpp",
+        "tuple": "cpp",
+        "type_traits": "cpp",
+        "utility": "cpp",
+        "fstream": "cpp",
+        "initializer_list": "cpp",
+        "iomanip": "cpp",
+        "istream": "cpp",
+        "limits": "cpp",
+        "new": "cpp",
+        "ostream": "cpp",
+        "sstream": "cpp",
+        "stdexcept": "cpp",
+        "streambuf": "cpp",
+        "typeinfo": "cpp",
+        "ios": "cpp",
+        "locale": "cpp",
+        "queue": "cpp",
+        "xfacet": "cpp",
+        "xhash": "cpp",
+        "xiosbase": "cpp",
+        "xlocale": "cpp",
+        "xlocbuf": "cpp",
+        "xlocinfo": "cpp",
+        "xlocmes": "cpp",
+        "xlocmon": "cpp",
+        "xlocnum": "cpp",
+        "xloctime": "cpp",
+        "xmemory": "cpp",
+        "xmemory0": "cpp",
+        "xstddef": "cpp",
+        "xtr1common": "cpp",
+        "xtree": "cpp",
+        "xutility": "cpp",
+        "complex": "cpp",
+        "set": "cpp",
+        "bitset": "cpp",
+        "codecvt": "cpp",
+        "condition_variable": "cpp",
+        "filesystem": "cpp",
+        "mutex": "cpp",
+        "thread": "cpp",
+        "__nullptr": "cpp",
+        "bit": "cpp",
+        "compare": "cpp",
+        "concepts": "cpp",
+        "stack": "cpp",
+        "stop_token": "cpp"
+    }
+}

TensorRT-Base/CMakeLists.txt

@@ -0,0 +1,76 @@
+# @file     - CMakeLists.txt
+# @author   - wdn ([email protected])
+# @brief    - 工程编译文件
+# 
+# @copyright Copyright (c) 2021, wdn. All rights reserved.
+
+cmake_minimum_required(VERSION 3.0)
+
+project("tensorrt_base")
+
+set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+set(CMAKE_CXX_EXTENSIONS OFF)
+
+set(EXECUTABLE_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/bin) 
+set(LIBRARY_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/lib)
+
+if(WIN32)
+  set(CUDA_ROOT_DIR "D:/Software/CUDA11.0/development")
+  set(TRT_ROOT_DIR "D:/Software/CUDA11.0/development")
+
+  # cuda
+  set(CUDA_INCLUDE_DIR ${CUDA_ROOT_DIR}/include)
+  set(CUDA_LIB_DIR ${CUDA_ROOT_DIR}/lib/x64)
+  file(GLOB CUDA_LIBRARIRS "${CUDA_ROOT_DIR}/lib/x64/*.lib")
+
+  # tensorrt
+  set(TRT_INCLUDE_DIR ${TRT_ROOT_DIR}/include)
+  set(TRT_LIB_DIR ${TRT_ROOT_DIR}/lib/x64)
+  file(GLOB TRT_LIBRARIRS "${TRT_ROOT_DIR}/lib/x64/*.lib")
+
+else()
+
+
+  set(CUDA_ROOT_DIR "/usr/local/cuda")
+  set(TRT_ROOT_DIR "/opt/TensorRT-7.2.2.3")
+
+    # cuda
+  set(CUDA_INCLUDE_DIR ${CUDA_ROOT_DIR}/include)
+  set(CUDA_LIB_DIR ${CUDA_ROOT_DIR}/lib64)
+  file(GLOB CUDA_LIBRARIRS "${CUDA_ROOT_DIR}/lib64/*.so")
+
+  # tensorrt
+  set(TRT_INCLUDE_DIR ${TRT_ROOT_DIR}/include)
+  set(TRT_LIB_DIR ${TRT_ROOT_DIR}/lib)
+  file(GLOB TRT_LIBRARIRS "${TRT_ROOT_DIR}/lib/*.so")
+
+endif()
+
+if(WIN32)
+  # Socket limit is 16K(can be raised arbitrarily)
+  add_definitions(-DFD_SETSIZE=16384)
+  add_definitions(-D_CRT_SECURE_NO_WARNINGS)
+  add_definitions(-D_WINSOCK_DEPRECATED_NO_WARNINGS)
+endif()
+
+# opencv
+find_package(OpenCV REQUIRED)
+message(STATUS "OpenCV library status:")
+message(STATUS "    version: ${OpenCV_VERSION}")
+message(STATUS "    libraries: ${OpenCV_LIBS}")
+message(STATUS "    include path: ${OpenCV_INCLUDE_DIRS}")
+
+include_directories(${CUDA_INCLUDE_DIR} ${TRT_INCLUDE_DIR} ${OpenCV_INCLUDE_DIRS})
+include_directories(common plugin)
+link_directories(${CUDA_LIB_DIR} ${TRT_LIB_DIR})
+
+# 包含子项目。
+add_subdirectory ("common")
+add_subdirectory ("plugin")
+add_subdirectory ("mnist_caffe")
+add_subdirectory ("mnist_onnx")
+add_subdirectory ("faster_rcnn")
+
+
+

TensorRT-Base/README.md

@@ -0,0 +1,113 @@
+# TensorRT Base
+
+## 1.Description
+该项目封装TensorRT加速深度学习模型，支持caffe和onnx格式模型。
+
+主要来自TensorRT C++ sample及部分Github参考代码。
+
+# 2.Environment
+
+推荐使用Ubuntu运行
+
+## 2.1 Ubuntu
+
+* TensorRT 7+
+* GUN工具（g++ 7.5.0）
+* CMake工具（3.10+）
+
+## 2.2 Windows
+
+* TensorRT 7+
+* Visual Studio 2017
+* CMake工具（3.10+）
+
+# 3.Directory
+
+```
+  ├── bin: 生成可执行文件目录
+  ├── build: CMake工具构建目录
+  ├── common: TensorRT基础类和一些共用函数，编译为库供调用
+  ├── lib: 生成库目录
+  ├── plugin: TensorRT plugin目录，用于存放自定义plugin
+  ├── CMakeLists.txt: 根目录CMake配置文件
+  ├── mnist_caffe: TensorRT解析caffe模型demo
+  ├── mnist_onnx: TensorRT解析ONNX模型demo
+  └── faster_rcnn: TensoRT加速Faster RCNN算法
+```
+
+# 4.Running
+
+修改根目录下CMakeLists.txt文件中`CUDA`和`TensorRT`目录，cudnn默认和CUDA安装在同一目录下。
+
+```cmake
+set(CUDA_ROOT_DIR "D:/Software/CUDA11.0/development")
+set(TRT_ROOT_DIR "D:/Software/CUDA11.0/development")
+```
+
+运行以下命令
+
+```shell
+cd TensorRT-Base
+mkdir build
+cd build
+cmake ..
+make
+```
+
+编译完成后，在bin文件夹中会生成可执行文件，lib目录下会生成库文件，由于Windows环境下配置动态库较麻烦，默认生成静态库，可自己修改。
+
+运行bin文件夹中可执行文件，即可加速深度学习模型，以faster rcnn demo为例
+
+```shell
+PS E:\TensorRT-Base> .\bin\Release\faster_rcnn.exe
+[2021-04-25 20:48:32][  WARN ] : Dynamic size input setting invalid!
+[2021-04-25 20:48:32][ DEBUG ] : deserialize engine from faster_rcnn/model/faster_rcnn.bin
+[2021-04-25 20:48:38][ DEBUG ] : max batch size of deserialized engine: 1
+[2021-04-25 20:48:38][ DEBUG ] : create execute context and malloc device memory...
+[2021-04-25 20:48:38][ DEBUG ] : init engine...
+[2021-04-25 20:48:38][ DEBUG ] : malloc device memory
+[2021-04-25 20:48:38][ DEBUG ] : nbBingdings: 5
+[2021-04-25 20:48:38][ DEBUG ] : input: 
+[2021-04-25 20:48:38][ DEBUG ] : binding bindIndex: 0, name: data, size in byte: 2250000
+[2021-04-25 20:48:38][ DEBUG ] : binding dims with 3 dimemsion
+[2021-04-25 20:48:38][ DEBUG ] : input: 
+[2021-04-25 20:48:38][ DEBUG ] : binding bindIndex: 1, name: im_info, size in byte: 12  
+[2021-04-25 20:48:38][ DEBUG ] : binding dims with 3 dimemsion
+[2021-04-25 20:48:38][ DEBUG ] : output: 
+[2021-04-25 20:48:38][ DEBUG ] : binding bindIndex: 2, name: rois, size in byte: 4800
+[2021-04-25 20:48:38][ DEBUG ] : binding dims with 3 dimemsion
+[2021-04-25 20:48:38][ DEBUG ] : output:
+[2021-04-25 20:48:38][ DEBUG ] : binding bindIndex: 3, name: bbox_pred, size in byte: 100800
+[2021-04-25 20:48:38][ DEBUG ] : binding dims with 4 dimemsion
+[2021-04-25 20:48:38][ DEBUG ] : output:
+[2021-04-25 20:48:38][ DEBUG ] : binding bindIndex: 4, name: cls_prob, size in byte: 25200
+[2021-04-25 20:48:38][ DEBUG ] : binding dims with 4 dimemsion
+[2021-04-25 20:48:38][  WARN ] : output_bbox_pred: 25200
+[2021-04-25 20:48:38][  WARN ] : output_cls_prob: 6300
+[2021-04-25 20:48:38][  WARN ] : output_rois: 1200
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 aeroplane
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 bicycle
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 bird
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 boat
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 bottle
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 bus
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 car
+[2021-04-25 20:48:38][  INFO ] : indices size is: 1 cat
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 chair
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 cow
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 diningtable
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 dog
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 horse
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 motorbike
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 person
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 pottedplant
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 sheep
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 sofa
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 train
+[2021-04-25 20:48:38][  INFO ] : indices size is: 0 tvmonitor
+```
+
+
+
+
+

TensorRT-Base/bin/README.md

@@ -0,0 +1 @@
+## 生成可执行文件目录

TensorRT-Base/common/CMakeLists.txt

@@ -0,0 +1,24 @@
+# @file         - common/CMakeLists.txt
+# @author       - wdn ([email protected])
+# @brief        - common编译文件
+# 
+# Copyright (c) 2021, wdn. All rights reserved.
+
+cmake_minimum_required (VERSION 3.0)
+
+project ("common")
+
+# aux_source_directory(. COMMON_SRC)
+file(GLOB COMMON_SOURCE *.cpp )
+
+# file(GLOB_RECURSE trt_source "helper.h" "helper.cpp" "logging.h" "timer.h")
+
+add_library(common STATIC ${COMMON_SOURCE})
+
+# 链接库
+target_link_libraries(
+        common PRIVATE 
+        trt_plugin
+        ${CUDA_LIBRARIRS} 
+        ${TRT_LIBRARIRS}
+        )