libtorch 1.12.1 cuda11.3 torch1.12.1 visual stdio2019环境搭建

conda 的环境

name: torch1.12.1
channels:
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
  - https://mirrors.ustc.edu.cn/anaconda/cloud/menpo/
  - https://mirrors.ustc.edu.cn/anaconda/cloud/bioconda/
  - https://mirrors.ustc.edu.cn/anaconda/cloud/msys2/
  - https://mirrors.ustc.edu.cn/anaconda/cloud/conda-forge/
  - https://mirrors.ustc.edu.cn/anaconda/pkgs/free/
  - https://mirrors.ustc.edu.cn/anaconda/pkgs/main/
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/msys2/
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/Paddle/
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/fastai/
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/bioconda/
dependencies:
  - blas=1.0=mkl
  - brotlipy=0.7.0=py310h2bbff1b_1002
  - bzip2=1.0.8=he774522_0
  - ca-certificates=2022.10.11=haa95532_0
  - certifi=2022.9.24=py310haa95532_0
  - cffi=1.15.1=py310h2bbff1b_0
  - charset-normalizer=2.0.4=pyhd3eb1b0_0
  - cryptography=38.0.1=py310h21b164f_0
  - cudatoolkit=11.3.1=h59b6b97_2
  - freetype=2.12.1=ha860e81_0
  - idna=3.4=py310haa95532_0
  - intel-openmp=2021.4.0=haa95532_3556
  - jpeg=9e=h2bbff1b_0
  - lerc=3.0=hd77b12b_0
  - libdeflate=1.8=h2bbff1b_5
  - libffi=3.4.2=hd77b12b_4
  - libpng=1.6.37=h2a8f88b_0
  - libtiff=4.4.0=h8a3f274_1
  - libuv=1.40.0=he774522_0
  - libwebp=1.2.4=h2bbff1b_0
  - libwebp-base=1.2.4=h2bbff1b_0
  - lz4-c=1.9.3=h2bbff1b_1
  - mkl=2021.4.0=haa95532_640
  - mkl-service=2.4.0=py310h2bbff1b_0
  - mkl_fft=1.3.1=py310ha0764ea_0
  - mkl_random=1.2.2=py310h4ed8f06_0
  - numpy=1.23.3=py310h60c9a35_0
  - numpy-base=1.23.3=py310h04254f7_0
  - openssl=1.1.1s=h2bbff1b_0
  - pillow=9.2.0=py310hdc2b20a_1
  - pip=22.2.2=py310haa95532_0
  - pycparser=2.21=pyhd3eb1b0_0
  - pyopenssl=22.0.0=pyhd3eb1b0_0
  - pysocks=1.7.1=py310haa95532_0
  - python=3.10.6=hbb2ffb3_1
  - pytorch=1.12.1=py3.10_cuda11.3_cudnn8_0
  - pytorch-mutex=1.0=cuda
  - requests=2.28.1=py310haa95532_0
  - setuptools=65.5.0=py310haa95532_0
  - six=1.16.0=pyhd3eb1b0_1
  - sqlite=3.39.3=h2bbff1b_0
  - tk=8.6.12=h2bbff1b_0
  - torchaudio=0.12.1=py310_cu113
  - torchvision=0.13.1=py310_cu113
  - typing_extensions=4.3.0=py310haa95532_0
  - tzdata=2022f=h04d1e81_0
  - urllib3=1.26.12=py310haa95532_0
  - vc=14.2=h21ff451_1
  - vs2015_runtime=14.27.29016=h5e58377_2
  - wheel=0.37.1=pyhd3eb1b0_0
  - win_inet_pton=1.1.0=py310haa95532_0
  - wincertstore=0.2=py310haa95532_2
  - xz=5.2.6=h8cc25b3_0
  - zlib=1.2.13=h8cc25b3_0
  - zstd=1.5.2=h19a0ad4_0
  - pip:
    - absl-py==1.3.0
    - albumentations==1.3.0
    - cachetools==5.2.0
    - colorama==0.4.6
    - contourpy==1.0.6
    - cycler==0.11.0
    - ensemble-boxes==1.0.9
    - ffmpeg==1.4
    - fonttools==4.38.0
    - google-auth==2.14.1
    - google-auth-oauthlib==0.4.6
    - grpcio==1.50.0
    - imageio==2.22.4
    - joblib==1.2.0
    - kiwisolver==1.4.4
    - llvmlite==0.39.1
    - markdown==3.4.1
    - markupsafe==2.1.1
    - matplotlib==3.6.2
    - networkx==2.8.8
    - numba==0.56.4
    - oauthlib==3.2.2
    - opencv-contrib-python==4.6.0.66
    - opencv-python==4.5.5.64
    - opencv-python-headless==4.6.0.66
    - packaging==21.3
    - pandas==1.5.1
    - protobuf==3.20.3
    - pyasn1==0.4.8
    - pyasn1-modules==0.2.8
    - pyparsing==3.0.9
    - python-dateutil==2.8.2
    - pytz==2022.6
    - pywavelets==1.4.1
    - pyyaml==6.0
    - qudida==0.0.4
    - requests-oauthlib==1.3.1
    - rsa==4.9
    - scikit-image==0.19.3
    - scikit-learn==1.1.3
    - scipy==1.9.3
    - seaborn==0.12.1
    - tensorboard==2.11.0
    - tensorboard-data-server==0.6.1
    - tensorboard-plugin-wit==1.8.1
    - thop==0.1.1-2209072238
    - threadpoolctl==3.1.0
    - tifffile==2022.10.10
    - tqdm==4.64.1
    - werkzeug==2.2.2
prefix: C:\Users\20169\.conda\envs\torch1.12.1

cuda环境

libtorch包 release版本的

void test_libtorch_version() {
	std::cout << "Hello Lbitorch:" << std::endl;
	std::cout << "	cuDNN available: " << torch::cuda::cudnn_is_available() << std::endl;
	std::cout << "	CUDA available: " << torch::cuda::is_available() << std::endl;
	std::cout << "  CUDA_VERSION: " << CUDA_VERSION << std::endl;
	std::cout << "  TORCH_VERSION: " << TORCH_VERSION << std::endl;
}

visual stdio2019配置
dll库拷贝到 x64/release里面，和生成的exe同一个目录

命令参数可以忽略
环境：

PATH=%PATH%
C:\Env\libTorch\lib

VC++目录

包含目录
C:\Env\opencv\buildCuda\install\include
C:\Env\opencv\buildCuda\install\include\opencv2
C:\Env\libtorch\include\torch\csrc\api\include
C:\Env\libtorch\lib
C:\Env\libtorch\include
C:\Env\linearAlgebra\eigen-3.4.0  #这个libtorch用不到可以忽略
库目录：
C:\Env\opencv\buildCuda\install\x64\vc16\lib
C:\Env\libtorch\lib

C/C++附加包含目录

C:\Env\libTorch\include
C:\Env\libTorch\include\torch\csrc\api\include

语言

链接器 —附加库目录

C:\Env\libTorch\lib

附加依赖项

C:\Env\libTorch\lib\torch_cuda.lib
C:\Env\libTorch\lib\torch_cpu.lib
C:\Env\libTorch\lib\asmjit.lib
C:\Env\libTorch\lib\c10.lib
C:\Env\libTorch\lib\c10_cuda.lib
C:\Env\libTorch\lib\clog.lib
C:\Env\libTorch\lib\cpuinfo.lib
C:\Env\libTorch\lib\dnnl.lib
C:\Env\libTorch\lib\fbgemm.lib
C:\Env\libTorch\lib\kineto.lib
C:\Env\libTorch\lib\libprotobuf.lib
C:\Env\libTorch\lib\torch.lib
C:\Env\libTorch\lib\torch_cuda_cpp.lib
C:\Env\libTorch\lib\torch_cuda_cu.lib
C:\Env\libTorch\lib\pthreadpool.lib
C:\Env\libTorch\lib\libprotobuf-lite.lib
C:\Env\libTorch\lib\caffe2_nvrtc.lib
C:\Env\libTorch\lib\XNNPACK.lib
opencv_world460.lib

如果cuda不可用，添加下面到连接器的命令行

/INCLUDE:?warp_size@cuda@at@@YAHXZ /INCLUDE:?_torch_cuda_cu_linker_symbol_op_cuda@native@at@@YA?AVTensor@2@AEBV32@@Z 

C++代码

#include<torch/torch.h>
#include<torch/script.h>
#include<iostream>

void test_libtorch_version() {
	std::cout << "Hello Lbitorch:" << std::endl;
	std::cout << "	cuDNN available: " << torch::cuda::cudnn_is_available() << std::endl;
	std::cout << "	CUDA available: " << torch::cuda::is_available() << std::endl;
	std::cout << "  TORCH_VERSION: " << TORCH_VERSION << std::endl;
}
auto loadModel(const std::string&& modelPath,torch::jit::Module & model,bool use_gpu=false) {
	if (torch::cuda::is_available() && use_gpu) {
		std::cout << "加载到CUDA上" << std::endl;
		auto device_type = torch::kCUDA;
		try {
			 model = torch::jit::load(modelPath, device_type);
			std::cout << "CUDA加载成功" << std::endl;
		}
		catch (const c10::Error& e) {
			std::cout<< "CUDA加载失败" << std::endl;
			std::exit(EXIT_FAILURE);
		}
	}
	else {
		std::cout << "加载到CPU上" << std::endl;
		auto device_type = torch::kCPU; //默认也是cpu
		try {
			model = torch::jit::load(modelPath, device_type);
			std::cout << "Cpu加载成功" << std::endl;
		}
		catch (const c10::Error& e) {
			std::cout << "Cpu加载失败" << std::endl;
			std::exit(EXIT_FAILURE);
		}
	}
}
int main() {
	test_libtorch_version();

	torch::Device device(torch::kCUDA);
	torch::Tensor tensor1 = torch::eye(3); // (A) tensor-cpu
	std::cout << tensor1 << std::endl;

	torch::Tensor tensor2 = torch::eye(3, device); // (B) tensor-cuda
	std::cout << "cuda .." << std::endl;
	std::cout << tensor2 << std::endl;
	auto your_path_cpu = "D:/pycharm/YOLO/xfs/yolov5-6.0/runs/train/exp/weights/ts_cpu.pt";
	auto your_path_gpu = "D:/pycharm/YOLO/xfs/yolov5-6.0/runs/train/exp/weights/ts_gpu.pt";
	
	torch::jit::Module model_c;
	std::cout << "加载cpu的PT模型" << std::endl;
	loadModel(your_path_cpu, model_c,false);
	
	torch::jit::Module model_g;
	std::cout << "加载gpu的PT模型" << std::endl;
	loadModel(your_path_gpu, model_g,true);
return 0;
}

python部分：
yolov5—> export.py

# YOLOv5