C++调用pytorch模型，libtorch ubuntu/windows配置

参考博客：https://blog.csdn.net/tlzhatao/article/details/86555269?depth_1-utm_source=distribute.pc_relevant.none-task&utm_source=distribute.pc_relevant.none-task

将PyTorch模型转化为Torch Script的两种方法

tracing

import torch
import torchvision

# An instance of your model.
model = torchvision.models.resnet18()

# model.eval()

# An example input you would normally provide to your model's forward() method.
example = torch.rand(1, 3, 224, 224)

# Use torch.jit.trace to generate a torch.jit.ScriptModule via tracing.
traced_script_module = torch.jit.trace(model, example)

# ScriptModule
output = traced_script_module(torch.ones(1, 3, 224, 224))

traced_script_module.save("resnet18_model.pt")

annotation

import torch
import torchvision

class MyModule(torch.jit.ScriptModule):
    def __init__(self):
        super(MyModule, self).__init__()
        self.weight = torch.nn.Parameter(torch.rand(1, 3, 224, 224))

    @torch.jit.script_method
    def forward(self, input):
        if input.sum() > 0:
          output = self.weight.mv(input)
        else:
          output = self.weight + input
        return output

my_script_module = MyModule()
my_script_module.save("model.pt")

ubuntu

官网下载对应的libtorch：https://download.pytorch.org/libtorch/nightly/cu100/libtorch-shared-with-deps-latest.zip，网页下载太慢可以用wget。
准备example-app.cpp文件

#include "torch/script.h"
#include "torch/torch.h"
#include <ctime>
#include <iostream>
#include <memory>

using namespace std;

int main(int argc, const char* argv[])
{
    if (argc != 2) {
        std::cerr << "usage: example-app <path-to-exported-script-module>\n";
        return -1;
    }

    // 读取我们的权重信息
    torch::jit::script::Module module = torch::jit::load(argv[1]);
    module.to(at::kCUDA);

    std::cout << "ok\n";

    // 建立一个输入，维度为(1,3,224,224)，并移动至cuda
    
    for(int i=0;i<100;i++){
        std::vector<torch::jit::IValue> inputs;
        inputs.push_back(torch::ones({1, 3, 224, 224}).to(at::kCUDA));
        clock_t startTime,endTime;
        startTime = clock();
        at::Tensor output = module.forward(inputs).toTensor();
        endTime = clock();//计时结束
        cout << "The run time is: " <<(double)(endTime - startTime) / CLOCKS_PER_SEC << "s" << endl;
        
    }

// Execute the model and turn its output into a tensor.
    

    //std::cout << output.slice(/*dim=*/1, /*start=*/0, /*end=*/5) << '\n';
}

准备CmakeLists.txt文件
注意最后一行标准是14，否则会报错。

cmake_minimum_required(VERSION 3.0 FATAL_ERROR)
project(example-app)
set(Torch_DIR /home/wxy/code/libtorch)
find_package(Torch REQUIRED)

add_executable(example-app example-app.cpp)
target_link_libraries(example-app "${TORCH_LIBRARIES}")
set_property(TARGET example-app PROPERTY CXX_STANDARD 14)

新建build文件

cmake -DCMAKE_PREFIX_PATH=/home/wxy/code/libtorch ..
make
./example-app ../model.pt

windows

官网下载对应的libtorch：https://download.pytorch.org/libtorch/nightly/cu90/libtorch-win-shared-with-deps-latest.zip，网页下载太慢可以用wget。
准备example-app.cpp文件
注意，如果是1.1版本及以下，是个指针，1.2以上已经改了，用到module的地方也相应改一下:

std::shared_ptr<torch::jit::script::Module> module = torch::jit::load(argv[1]);

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

using namespace std;
 
int main(int argc, const char* argv[]) {
  /*if (argc != 2) {
    std::cerr << "usage: example-app <path-to-exported-script-module>\n";
    return -1;
  }*/
 
  // Deserialize the ScriptModule from a file using torch::jit::load().
  torch::jit::script::Module module = torch::jit::load("F:/python_project/pytorch/inference_c++/demo/resnet18_model.pt");
  module.to(at::kCUDA);
  std::cout << "ok\n";

    // Create a vector of inputs.
  std::vector<torch::jit::IValue> inputs;
  inputs.push_back(torch::rand({1, 3, 224, 224}).to(at::kCUDA));
  
  // Execute the model and turn its output into a tensor.
  auto output = module.forward(inputs).toTensor();
  
  std::cout << output.slice(/*dim=*/1, /*start=*/0, /*end=*/10) << '\n';
}

准备CmakeLists.txt文件
注意最后一行标准是14，否则会报错。

cmake_minimum_required(VERSION 3.0 FATAL_ERROR)
project(example-app)
set(Torch_DIR F:/python_project/pytorch/inference_c++/libtorch-win-shared-with-deps-latest/libtorch)
find_package(Torch REQUIRED)
add_executable(example-app example-app.cpp)
target_link_libraries(example-app “${TORCH_LIBRARIES}”)
set_property(TARGET example-app PROPERTY CXX_STANDARD 14)

新建build文件

cmake -DCMAKE_PREFIX_PATH=F:/python_project/pytorch/inference_c++/libtorch-win-shared-with-deps-latest/libtorch -DCMAKE_GENERATOR_PLATFORM=x64 -DCMAKE_BUILD_TYPE=Release ..
make

最后用VS打开工程，因为我的cuda路径不是默认的路径，所以有一些库目录可能需要改一下，最后生成.exe文件运行可能需要nvToolsExt64_1.dll，在网上下载一个就可以了，这里提供一个链接：https://www.dll4free.com/nvtoolsext64_1.dll.html