Include could not find load file: /path/build/TorchVisionTargets.cmake

vision
1

Hi,
I am trying to run vision c++ api with torch. I have added the following in in my CMakeLists.txt.

find_package(TorchVision REQUIRED)
target_link_libraries(my-target PUBLIC TorchVision::TorchVision)

But when call cmake I get:

ubuntu@pc:~/pytorch_cpp/build$ cmake -DCMAKE_PREFIX_PATH="/home/ubuntu/pytorch_cpp/build/libtorch;/home/ubuntu/pytorch_cpp/build/vision" ..
-- Caffe2: CUDA detected: 10.0
-- Caffe2: CUDA nvcc is: /usr/local/cuda/bin/nvcc
-- Caffe2: CUDA toolkit directory: /usr/local/cuda
-- Caffe2: Header version is: 10.0
-- Found cuDNN: v7.5.1  (include: /usr/local/cuda/include, library: /usr/local/cuda/lib64/libcudnn.so)
-- Autodetected CUDA architecture(s):  5.2 5.2
-- Added CUDA NVCC flags for: -gencode;arch=compute_52,code=sm_52
CMake Error at build/vision/build/TorchVisionConfig.cmake:50 (include):
  include could not find load file:

    /home/ubuntu/pytorch_cpp/build/vision/build/TorchVisionTargets.cmake
Call Stack (most recent call first):
  CMakeLists.txt:10 (find_package)


CMake Error at build/vision/build/TorchVisionConfig.cmake:59 (set_target_properties):
  set_target_properties Can not find target to add properties to:
  TorchVision::TorchVision
Call Stack (most recent call first):
  CMakeLists.txt:10 (find_package)


-- Configuring incomplete, errors occurred!
See also "/home/ubuntu/pytorch_cpp/build/CMakeFiles/CMakeOutput.log".
See also "/home/ubuntu/pytorch_cpp/build/CMakeFiles/CMakeError.log".

I checked /home/ubuntu/pytorch_cpp/build/vision/build/ for TorchVisionTargets.cmake and its not present. I called cmake in the directory, but it was not created. I’m not sure how to go around this issue as I’m pretty new to LibTorch. I’m trying to run the RCNN-FPN model from detectron2 on torchscript. If I run it without vision I get the error:
Unknown builtin op: torchvision::nms. Could not find any similar ops to torchvision::nms. This op may not exist or may not be currently supported in TorchScript.

I need some guidance on how to set up vision in c++. There are my files:

CMakeLists.txt

cmake_minimum_required(VERSION 3.10 FATAL_ERROR)
project(example)
#set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CXX_STANDARD 14)
find_package(Torch REQUIRED)


add_executable(main main.cpp)
target_link_libraries(main ${TORCH_LIBRARIES})

main.cpp

#include <torch/torch.h>
#include <torch/script.h>
#include <torchvision/vision.h>

int main()
{
    torch::Device device(torch::kCPU);
    torch::Tensor tensor = torch::zeros({2, 2});
    std::cout << tensor << std::endl;


    if (torch::cuda::is_available()) {
        std::cout << "CUDA is available! " << std::endl;
        device = torch::kCUDA;
    }

    torch::Tensor test_gpu_tensor = tensor.to(device);

    std::cout << test_gpu_tensor << std::endl;


    torch::jit::script::Module module;
    module = torch::jit::load("/home/ubuntu/pytorch_cpp/script_model.pt");
    module.to(device);
    return 0;
}

Directory Structure:

├── CMakeLists.txt
├── build
│   ├── CMakeCache.txt
│   ├── CMakeFiles [14 entries exceeds filelimit, not opening dir]
│   ├── Makefile
│   ├── cmake_install.cmake
│   ├── detect_cuda_compute_capabilities.cpp
│   ├── detect_cuda_version.cc
│   ├── libtorch
│   │   ├── bin
│   │   ├── build-hash
│   │   ├── build-version
│   │   ├── include
│   │   ├── lib
│   │   └── share
│   ├── main
│   └── vision
          └── build
├── main.cpp
└── script_model.pt

References:

2

Hello, I have the same problem with you, and I have solved this problem right now.
The torchvisionTarget.cmake is in the CMakeFiles/Export/[some random code]/. But I don’t know why this file is in that place. And my solution does not need it.
Since you have built the torchvision successfully, you can easily find the torchvision.lib and torchvision.dll in the build/. And you can just link these two file into your project like:

  1. Add include_directories(path/to/vision-project/torchvision/csrc) so that you can find the torchvision header easily in your project.
  2. Add ‘target_link_libraries(your_target path/to/your/torchvision.lib)’
    And then build your project. You also need to move the ‘torchvision.dll’ to the directory where your target executable file placed.

I’m also a noob in C++, and this problem has confused my for a few days. This solution can pass the C++ example in vision. Hope these information can help you.