opencv+libtorch c++工程问题及解决方案

问题1：添加opencv库后，报错：

 by not providing "findqt5core.cmake" in cmake_module_path this project has  asked cmake to find a ...

• 原因：未安装qt5
• 解决方案：安装qt5

sudo apt-get install qt5-default

安装后编译时，仍然报错如上，解决方案：安装QT5后重启电脑。

问题2：opencv报错

没有规则可制作目标“opencv/build/lib/libopencv_gapi.so.4.5.2libopencv_gapi.so.4.5.2”

• 原因：只采用cmake-gui对opencv进行configure和Generate，尚未进行编译和安装
• 解决方案：在build中打开终端：

make
make install

问题3：CMakeLists设置

cmake_minimum_required(VERSION 3.17)

project(enhance)
set(CMAKE_INCLUDE_CURRENT_DIR ON)
set(CMAKE_AUTOMOC ON)
set(BOOST_LIBRARYDIR "/usr/lib/x86_64-linux-gnu")
find_package(Qt5Core)

add_executable(${PROJECT_NAME} "main.cpp" xx.h xx.cpp xx.cpp)

target_link_libraries(${PROJECT_NAME} Qt5::Core)


## torch
#set(CMAKE_PREFIX_PATH /media/zyy/software/libtorch/share/cmake/Torch)
set(CMAKE_PREFIX_PATH /media/zyy/software/anaconda3/envs/pytorch160/lib/python3.7/site-packages/torch)
find_package(Torch REQUIRED NO_CMAKE_FIND_ROOT_PATH)
target_link_libraries(${PROJECT_NAME} "${TORCH_LIBRARIES}")

# OpenCV
find_package(OpenCV REQUIRED)
# 添加头文件
include_directories(${OpenCV_INCLUDE_DIRS})
# Link your application with OpenCV libraries
target_link_libraries(${PROJECT_NAME} ${OpenCV_LIBS})

问题4：undefined reference to “cv::imread”

现象：opencv和torch分别使用不会报错，但一起使用报上述找不到cv中的函数

原因： opencv4.x编译后与libtorch有冲突
解决方案： 编译安装opencv3.4.14版本（opencv3.x应该都可行）

• 卸载安装的opencv:

sudo make uninstall
cd ..
sudo rm -r build
sudo rm -r /usr/local/include/opencv2 /usr/local/include/opencv /usr/include/opencv /usr/include/opencv2 /usr/local/share/opencv /usr/local/share/OpenCV /usr/share/opencv /usr/share/OpenCV /usr/local/bin/opencv* /usr/local/lib/libopencv*
sudo apt-get --purge remove opencv-doc opencv-data python-opencv

• 检查

locate opencv

• 重新编译安装opencv3.4.14（网上有很多教程，请自行编译安装）

问题5：Mat转Tensor后放入GPU报错

Process finished with exit code 11

源代码如下：

torch::Tensor input_tensor = torch::from_blob(img.data, {1, img_float.rows, img_float.cols, img.channels()});
input_tensor = input_tensor.toType(torch::kFloat).to(torch::kCUDA);

猜测原因：tensor和Mat类型应对应，Mat为CV_8U，解决方案有二;
解决方案1（缺点耗时）：

cv::Mat img_float;
img.convertTo(img_float, CV_32FC3);
auto img_tensor = torch::from_blob(img_float.data, { 1, img_float.rows, img_float.cols, img.channels() }).to(torch::kCUDA);
img_tensor = img_tensor.permute({0, 3, 1, 2});

解决方案2：

auto img_tensor = torch::from_blob(img.data, { 1, img.rows, img.cols, img.channels()}, torch::kByte).to(torch::kCUDA);
img_tensor = img_tensor.permute({0, 3, 1, 2}).toType(torch::kFloat);

问题6：libtorch GPU占用大，导致GPU溢出

在python环境中，模型推理之前加入 with torch.no_grad(), 可以做计算图，减少GPU占用

• 解决方案：

torch::NoGradGuard no_grad;
torch::Tensor output = model.forward(inputs).toTensor();