win10+vs2017+denseflow编译

由于要训练自己动作识别模型，所以需要制作相关数据集，其中就是要制作光流数据，尝试使用传统opencv来做，发现速度太慢。最后找到denseflow库，于是在此记录在win10下的安装过程。

一、安装依赖

1.1、CUDA

• CUDA (driver version > 400)

这里默认已经安装，搞深度学习安装pyTorch/Tensorflow应该都安装过吧。这里我的cuda环境是：CUDA10.0

C:\Users\Administrator>nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:04_Central_Daylight_Time_2018
Cuda compilation tools, release 10.0, V10.0.130

1.2、OpenCV

• OpenCV (with CUDA support)

这里需要安装支持CUDA的OpenCV,博主在这里卡了好久。。。，这里需要在编译OpenCV时注意以下选项：

• BUILD_CUDA_STUBS:选中
• OPENCV_DNN_CUDA:未选中
• WITH_CUDA:选中

同时，主要这里需要opencv_contrib模块，需要和OpenCV一起安装

在选择编译器时，最好选择vs2017或vs2015，其他的可能编译不成功。

安装可以参考以下链接：VS2019+opencv4.5+CMake编译opencv_contrib-4.5.0模块（较为详细教程）

1.3、Boost

• Boost

正常安装，网上也有很多相关教程，如：Windows10下配置Boost

二、Cmake编译

2.1、生成可编译文件sln

这里，我们可以参考github上教程直接执行：

git clone https://github.com/open-mmlab/denseflow.git
mkdir build && cd build
cmake -DCMAKE_INSTALL_PREFIX=$HOME/app -DUSE_HDF5=no -DUSE_NVFLOW=no ..

也可以在clone源代码后，使用cmake-gui进行编译,这里我选择第一种方法了。所以，我们需要改相关CMakeLists.txt文件内容，修改如下：

• 添加OpenCV库

这一步主要是为了让Cmake发现OpenCV的位置，修改如下：

set(OpenCV_DIR D:\\cplusplus\\library\\opencv\\opencv-4.5.2\\build)

具体需要参考你的OpenCV编译的位置。

• 添加Boost库

set(Boost_USE_STATIC_LIBS ON)
set(BOOST_ROOT D:\\cplusplus\\library\\boost_1_73_0)

• 删除pthread依赖

target_link_libraries(${DenseFlow_LIB} ${OpenCV_LIBS} Boost::filesystem ${Boost_LIBRARIES})

完整CMakeLists.txt文件如下：

cmake_minimum_required(VERSION 3.12)
project(DenseFlow)

option(USE_HDF5 "Whether to build hdf5 for optical flow image saving" OFF)
option(USE_NVFLOW "Whether to use nvidia hardware optical flow" OFF)

set(OpenCV_DIR D:\\cplusplus\\library\\opencv\\opencv-4.5.2\\build)
set(Boost_USE_STATIC_LIBS ON)
set(BOOST_ROOT D:\\cplusplus\\library\\boost_1_73_0)

include(CheckCXXCompilerFlag)
CHECK_CXX_COMPILER_FLAG("-std=c++11" COMPILER_SUPPORTS_CXX11)
CHECK_CXX_COMPILER_FLAG("-std=c++0x" COMPILER_SUPPORTS_CXX0X)
if(COMPILER_SUPPORTS_CXX11)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC")
elseif(COMPILER_SUPPORTS_CXX0X)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++0x")
else()
    message(STATUS "The compiler ${CMAKE_CXX_COMPILER} has no C++11 support. Please use a different C++ compiler.")
endif()

find_package(CUDA REQUIRED)
find_package(OpenCV REQUIRED)

set(DenseFlow_LIB zzdenseflow)

message(STATUS "OpenCV library status:")
message(STATUS "    config: ${OpenCV_DIR}")
message(STATUS "    version: ${OpenCV_VERSION}")
message(STATUS "    libraries: ${OpenCV_LIBS}")
message(STATUS "    include path: ${OpenCV_INCLUDE_DIRS}")

find_package(Boost REQUIRED COMPONENTS date_time filesystem iostreams)
message(STATUS "Boost library status:")
message(STATUS "    version: ${Boost_VERSION}")
message(STATUS "    libraries: ${Boost_LIBRARIES}")
message(STATUS "    include path: ${Boost_INCLUDE_DIRS}")

add_library(${DenseFlow_LIB} src/common.cpp src/utils.cpp src/denseflow_gpu.cpp)
include_directories(${OpenCV_INCLUDE_DIRS} ${Boost_INCLUDE_DIRS} include ${PROJECT_BINARY_DIR})
link_directories(${OpenCV_LIB_DIRS})
target_link_libraries(${DenseFlow_LIB} ${OpenCV_LIBS} Boost::filesystem ${Boost_LIBRARIES})

if (USE_HDF5)
    find_package(HDF5 COMPONENTS HL REQUIRED)
    message(STATUS "HDF5 library status:")
    message(STATUS "    version: ${HDF5_VERSION}")
    message(STATUS "    libraries: ${HDF5_LIBRARIES} ${HDF5_HL_LIBRARIES}")
    message(STATUS "    include path: ${HDF5_INCLUDE_DIRS} ${HDF5_HL_INCLUDE_DIR}")
    set(HDF5_DEFINITION 1)
    include_directories(${HDF5_INCLUDE_DIRS} ${HDF5_HL_INCLUDE_DIR})
    target_link_libraries(${DenseFlow_LIB} ${HDF5_LIBRARIES} ${HDF5_HL_LIBRARIES})
else()
    set(HDF5_DEFINITION 0)
endif()

if (USE_NVFLOW)
    set(NVFLOW_DEFINITION 1)
else()
    set(NVFLOW_DEFINITION 0)
endif()

configure_file(
    "${PROJECT_SOURCE_DIR}/include/config.h.in"
    "${PROJECT_BINARY_DIR}/config.h"
)

add_executable(denseflow tools/denseflow.cpp)
target_link_libraries(denseflow ${DenseFlow_LIB})

install (
    TARGETS denseflow ${DenseFlow_LIB}
    RUNTIME DESTINATION bin
    ARCHIVE DESTINATION lib
    LIBRARY DESTINATION lib
)

执行完成后可以发现在源代码下有.sln文件，打开就可以进行vs编译了。

2.2、编译生成denseflow.exe

点击打开vs2017,进行编译。会出现以下错误：

• error C2872: “ACCESS_MASK”: 不明确的符号

原因：include/common.h中using namespace cv;造成的空间冲突造成的

解决方法：删除include/common.h中的using namespace cv;空间命名，同时，需要在用到OpenCV中函数前面添加cv：：命名空间

如：void convertFlowToImage(const Mat &flow_x, const Mat &flow_y, Mat &img_x, Mat &img_y, double lowerBound,double higherBound);

改为void convertFlowToImage(const cv::Mat &flow_x, const cv::Mat &flow_y, cv::Mat &img_x, cv::Mat &img_y, double lowerBound,double higherBound);

其他地方也与之类似。这需要把所有相关代码都改了，不然会报错。

• fatal error C1083: 无法打开包括文件: “sys/syscall.h”: No such file or directory

原因：这个头文件时linux上的，如果在windwos上的话，需要改为windows.h头文件

解决方法：在src/denseflow_gpu.cpp注销#include <sys/syscall.h>,添加以下内容：

//#include <sys/syscall.h>
#if __linux
#include <sys/syscall.h>
#elif defined(_WIN32) || defined(_WIN64)
#include <windows.h>       // Or something like it. 
#endif

最终，我们编译成功。

在build/Release文件夹下发现我们生成的denseflow.exe。