安装libtorch记录

安装libtorch记录

环境

• win10 + vs2017 + pytorch1.6.0 + cmake3.12.1 + opencv3.4.5 + cuda10.2

import torch as t 
print(t.version.cuda)
print(t.__version__)

10.2
1.6.0

步骤

1. 下载对应版本的libtorch

2. 创建工程文件夹，包含build文件夹、example-app.cpp、CMakeLists.txt，使用cmake创建VS工程，需要设置对opencv和libtorch路径

cmake_minimum_required(VERSION 3.12 FATAL_ERROR)
project(example-app)
 
set(CMAKE_PREFIX_PATH "D:\\Program Files (x86)\\libtorch\\share\\cmake\\Torch")

find_package(Torch REQUIRED)
find_package(OpenCV REQUIRED)

if(NOT Torch_FOUND)
    message(FATAL_ERROR "Pytorch Not Found!")
endif(NOT Torch_FOUND)

message(STATUS "Pytorch status:")
message(STATUS "    libraries: ${TORCH_LIBRARIES}")

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


add_executable(example-app example-app.cpp)
target_link_libraries(example-app ${TORCH_LIBRARIES} ${OpenCV_LIBS})
set_property(TARGET example-app PROPERTY CXX_STANDARD 11)

3. 打开build文件夹下的vs工程，将example-app设为启动项

验证

转换权重

import torch as t
from minist_experiment.models import LeNet

model_save_path = './checkpoints/checkpoints_epoch99.pth'

model = LeNet()
model.load_state_dict(t.load(model_save_path))
model.eval()

example = t.rand(1,1,28,28)

traced_script_module = t.jit.trace(model, example)
output = traced_script_module(t.ones(1, 1, 28, 28))
print(output.data)
traced_script_module.save("./checkpoints/checkpoints_Script.pt")

C++中读取模型

#include <memory>
#include <torch/torch.h>
#include <torch/script.h> // One-stop header.


int main() {
	// Deserialize the ScriptModule from a file using torch::jit::load().
	torch::jit::script::Module module = torch::jit::load("D:/Code/python/pytorch/minist_experiment/checkpoints/checkpoints_Script.pt");
	//assert(module != nullptr);
	std::cout << "ok\n";
	 // Create a vector of inputs.
	std::vector<torch::jit::IValue> inputs;
	inputs.push_back(torch::ones({ 1, 1, 28, 28 }));

  // Execute the model and turn its output into a tensor.
	torch::Tensor output = module.forward(inputs).toTensor();

	 std::cout << output.slice(/*dim=*/1, /*start=*/0, /*end=*/10) << '\n';
	 while (1);

}

遇到的问题

1. vs报错

error C2210: “_Ty”: 包扩展不能被用作别名模板中非打包参数的自变量

• 问题原因：C compiler版本太老
• 解决办法：更新一下VS

2. 运行代码报错VCRUNTIME140_1D.dll找不到

• 解决办法：在https://www.dll-files.com/中下载缺少的dll文件，放到C:\Windows\SysWOW64下

3. 缺少torch.dll等dll文件

• 解决办法：在libtorch安装路径里的lib（D:\Program Files (x86)\libtorch\lib）里找到对应文件添加到你工程文件夹下的这个目录下D:\Program Files (x86)\libtorchproject\build\x64\Debug

过程中参考了网上很多大哥的博客，这里就不一一列举了