c++ 使用torchscript 加载训练好的pytorch模型

1.首先官网上下载libtorch，放到当前项目下

2.将pytorch训练好的模型使用torch.jit.trace导出为.pt格式

 import torch
 from skimage import io, transform, color
 import numpy as np
 import os
 import torch.nn.functional as F
 import warnings
 warnings.filterwarnings("ignore")
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
 labels = ['cock', 'drawing', 'neutral', 'porn', 'sexy']
 path = "test/n_1.jpg"
 im = io.imread(path)
 if im.shape[2] == 4:
     im = color.rgba2rgb(im)
 
 im = transform.resize(im, (224, 224))
 im = np.transpose(im, (2, 0, 1))
 dummy_input = np.expand_dims(im, 0)
 inp = torch.from_numpy(dummy_input)
 inp = inp.float()
 model = torch.load(
     "models/resnet50-epoch-0-accu-0.9213857428381079.pth", map_location='cpu')
 traced_script_module = torch.jit.trace(model, inp)
 output = model(inp)
 probs = F.softmax(output).detach().numpy()[0]
 pred = np.argmax(probs)
 
 traced_script_module.save("models/traced_resnet_model.pt")

torchscript加载.pt模型

 // One-stop header.
 #include <torch/script.h>
 
 // headers for opencv
 #include <opencv2/highgui/highgui.hpp>
 #include <opencv2/imgproc/imgproc.hpp>
 #include <opencv2/opencv.hpp>
 
 #include <cmath>
 #include <iostream>
 #include <memory>
 #include <string>
 #include <vector>
 
 #define kIMAGE_SIZE 224
 #define kCHANNELS 3
 #define kTOP_K 1 //print top k predicted results
 
 bool LoadImage(std::string file_name, cv::Mat &image)
 {
   image = cv::imread(file_name); // CV_8UC3
   if (image.empty() || !image.data)
   {
     return false;
   }
   cv::cvtColor(image, image, CV_BGR2RGB);
   // scale image to fit
   cv::Size scale(kIMAGE_SIZE, kIMAGE_SIZE);
   cv::resize(image, image, scale);
 
   // convert [unsigned int] to [float]
   image.convertTo(image, CV_32FC3,1.0/255);
 
   return true;
 }
 
 bool LoadImageNetLabel(std::string file_name,
                        std::vector<std::string> &labels)
 {
   std::ifstream ifs(file_name);
   if (!ifs)
   {
     return false;
   }
   std::string line;
   while (std::getline(ifs, line))
   {
     labels.push_back(line);
   }
   return true;
 }
 
 int main(int argc, const char *argv[])
 {
   if (argc != 3)
   {
     std::cerr << "Usage:classifier <path-to-exported-script-module>  <path-to-lable-file> " << std::endl;
     return -1;
   }
 
   //load model
   torch::jit::script::Module module = torch::jit::load(argv[1]);
   // to GPU
   // module->to(at::kCUDA);
   std::cout << "== ResNet50 loaded!\n";
 
   //load labels(classes names)
   std::vector<std::string> labels;
   if (LoadImageNetLabel(argv[2], labels))
   {
     std::cout << "== Label loaded! Let's try it\n";
   }
   else
   {
     std::cerr << "Please check your label file path." << std::endl;
     return -1;
   }
 
   std::string file_name = "";
   cv::Mat image;
   while (true)
   {
     std::cout << "== Input image path: [enter q to exit]" << std::endl;
     std::cin >> file_name;
     if (file_name == "Q" || file_name == "q")
     {
       break;
     }
     if (LoadImage(file_name, image))
     {
       //read image tensor
       auto input_tensor = torch::from_blob(
           image.data, {1, kIMAGE_SIZE, kIMAGE_SIZE, kCHANNELS});
       input_tensor = input_tensor.permute({0, 3, 1, 2});
       input_tensor[0][0] = input_tensor[0][0].sub_(0.485).div_(0.229);
       input_tensor[0][1] = input_tensor[0][1].sub_(0.456).div_(0.224);
       input_tensor[0][2] = input_tensor[0][2].sub_(0.406).div_(0.225);
       // to GPU
       // input_tensor = input_tensor.to(at::kCUDA);
 
       torch::Tensor out_tensor = module.forward({input_tensor}).toTensor();
 
       auto results = out_tensor.sort(-1, true);
       auto softmaxs = std::get<0>(results)[0].softmax(0);
       auto indexs = std::get<1>(results)[0];
 
       for (int i = 0; i < kTOP_K; ++i)
       {
         auto idx = indexs[i].item<int>();
         std::cout << "    ============= Top-" << i + 1 << " =============" << std::endl;
         std::cout << "    Label:  " << labels[idx] << std::endl;
         std::cout << "    With Probability:  "
                   << softmaxs[i].item<float>() * 100.0f << "%" << std::endl;
       }
     }
     else
     {
       std::cout << "Can't load the image, please check your path." << std::endl;
     }
   }
 }

CMakeLists.txt编译

 cmake_minimum_required(VERSION 2.8)
 project(predict_demo)
 SET(CMAKE_CXX_FLAGS ${CMAKE_CXX_FLAGS} "-std=c++11 -O3")
 
 set(OpenCV_DIR  /home/buyizhiyou/opencv-3.4./build)
 find_package(OpenCV REQUIRED)
 find_package(Torch REQUIRED)
 
 # 添加头文件
 include_directories( ${OpenCV_INCLUDE_DIRS} )
 
 add_executable(resnet_demo resnet_demo.cpp)
 target_link_libraries(resnet_demo ${TORCH_LIBRARIES}  ${OpenCV_LIBS})
 set_property(TARGET resnet_demo PROPERTY CXX_STANDARD )

运行

./resnet_demo   models/traced_resnet_model.pt  labels.txt