C++：利用Libtorch调用保存的pytorch模型（.pt）进行前向计算（预测）

谁践踏了优雅 2023-09-25 13:01 10阅读 0赞

## 一、Pytorch框架下生成pt模型（python） ##

**python脚本**

# pth转pt
    import os
    import torch
    from PIL import Image
    from torchvision import transforms
    from model import AlexNet
    def main():
        device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
        print("using {} device.".format(device))
        # create model
        model = AlexNet(num_classes=5).to(device)
        img_path = r'D:\项目\人脸识别\deep-learning-for-image-processing-master\data_set\flower_data\rose.jpg'
        image = Image.open(img_path).convert('RGB')
        data_transform = transforms.Compose(
            [transforms.Resize((224, 224)),
             transforms.ToTensor(),
             transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
        img = data_transform(image)
        img = img.unsqueeze(dim=0)
        print(img.shape)
        # load model weights
        weights_path = "./AlexNet.pth"
        assert os.path.exists(weights_path), "file: '{}' dose not exist.".format(weights_path)
        testsize = 224
    
        if torch.cuda.is_available():
            modelState = torch.load(weights_path, map_location='cuda')
            model.load_state_dict(modelState, strict=False)
            model = model.cuda()
            model = model.eval()
            # An example input you would normally provide to your model's forward() method.
            example = torch.rand(1, 3, testsize, testsize)
            example = example.cuda()
            traced_script_module = torch.jit.trace(model, example)
            output = traced_script_module(img.cuda())
            print(output.shape)
            pred = torch.argmax(output, dim=1)
            print(pred)
    
            traced_script_module.save('model_cuda.pt')
        else:
            modelState = torch.load(weights_path, map_location='cpu')
            model.load_state_dict(modelState, strict=False)
            example = torch.rand(1, 3, testsize, testsize)
            example = example.cpu()
            traced_script_module = torch.jit.trace(model, example)
            output = traced_script_module(img.cpu())
            print(output.shape)
            pred = torch.argmax(output, dim=1)
            print(pred)
            traced_script_module.save('AlexNet.pt')
    if __name__ == '__main__':
        main()

## 二、C++环境下调用pt模型 ##

### 1、关键代码 ###

引入头文件

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

引入命名空间

using torch::jit::script::Module;

创建Libtorch模型对象

Module* module = new Module();

加载pt模型

*module = torch::jit::load("D:/项目/人脸识别/deep-learning-for-image-processing-master/pytorch_classification/Test2_alexnet/AlexNet.pt");

前向计算

at::Tensor output = module.forward({ tensor_image }).toTensor();

完整代码：

#include<iostream>
    #include <torch/script.h>
    #include "torch/torch.h"
    #include <opencv2/opencv.hpp>
    #include "opencv2/core.hpp"
    #include "opencv2/imgproc.hpp"
    #include "opencv2/highgui.hpp"
    #include "opencv2/imgcodecs.hpp"
    #include <vector>
    #include <chrono>
    #include <string>
    #include <vector>
    #include <memory>
    
    std::string classList[5] = { "daisy", "dandelion", "rose", "sunflower", "tulip" };
    std::string image_path = "D:/项目/人脸识别/deep-learning-for-image-processing-master/data_set/flower_data/tulip.jpg";
    
    int main(int argc, const char* argv[]) {
    	// Deserialize the ScriptModule from a file using torch::jit::load().
    	//std::shared_ptr<torch::jit::script::Module> module = torch::jit::load("../../model_resnet_jit.pt");
    	using torch::jit::script::Module;
    	Module module = torch::jit::load("D:/项目/人脸识别/deep-learning-for-image-processing-master/pytorch_classification/Test2_alexnet/AlexNet.pt");
    	std::cout << "测试图片：" << image_path << std::endl;
    	std::cout << "cuda support:" << (torch::cuda::is_available() ? "ture" : "false") << std::endl;
    	std::cout << "CUDNN:  " << torch::cuda::cudnn_is_available() << std::endl;
    	std::cout << "GPU(s): " << torch::cuda::device_count() << std::endl;
    	// module.to(at::kCUDA); //cpu下会在(auto image = cv::imread(image_path, cv::IMREAD_COLOR))行引起c10：error，未经处理的异常
    	module.eval();
    	module.to(at::kCPU);
    	//assert(module != nullptr);
    	//std::cout << "ok\n";
    	//输入图像
    	auto image = cv::imread(image_path, cv::IMREAD_COLOR);
    	cv::cvtColor(image, image, cv::COLOR_BGR2RGB);
    	cv::Mat image_transfomed = cv::Mat(cv::Size(224, 224), image.type());
    	cv::resize(image, image_transfomed, cv::Size(224, 224));
    	//cv::cvtColor(image_transfomed, image_transfomed, cv::COLOR_BGR2RGB);
    	// 转换为Tensor
    	torch::Tensor tensor_image = torch::from_blob(image_transfomed.data,
    		{ image_transfomed.rows, image_transfomed.cols,3 }, torch::kByte);
    	// 通道转换 {0,1,2} --> {2,0,1} {C,H,W} --> {H,W,C}
    	tensor_image = tensor_image.permute({ 2,0,1 });
    	// 数据类型转换
    	tensor_image = tensor_image.toType(torch::kFloat);
    	auto tensor_image_Tmp = torch::autograd::make_variable(tensor_image, false);
    	// 归一化，做除法：1/255
    	tensor_image = tensor_image.div(255);
    	// 标准化
    	tensor_image[0] = (tensor_image[0] - 0.5) / 0.5;
    	tensor_image[1] = (tensor_image[1] - 0.5) / 0.5;
    	tensor_image[2] = (tensor_image[2] - 0.5) / 0.5;
    	// 解压，将图片解压到1维
    	tensor_image = tensor_image.unsqueeze(0);
    	// tensor_image = tensor_image.to(at::kCUDA);
    	tensor_image = tensor_image.to(at::kCPU);
    	// 网络前向计算
    	at::Tensor output = module.forward({ tensor_image }).toTensor();
    	std::cout << "output:" << output << std::endl;
    	auto prediction = output.argmax(1);
    	std::cout << "prediction:" << prediction << std::endl;
    	int maxk = 5;
    	auto top3 = std::get<1>(output.topk(maxk, 1, true, true));
    	std::cout << "top3: " << top3 << '\n';
    	std::vector<int> res;
    	for (auto i = 0; i < maxk; i++) {
    		res.push_back(top3[0][i].item().toInt());
    	}
    	int pre = torch::Tensor(prediction).item<int>();
    	std::string result = classList[pre];
    	std::cout << "This is:" << result << std::endl;
    
    	cvWaitKey();
    	return 0;
    }

## 注意事项 —— 图像预处理 ##

在pytorch的.pth网络模型转.pt过程中，要保证图像image的预处理方式与之前在训练.pth模型时保持一致。

![f1e3e9cf5c884fdaa9c5178d9b33affc.png][]

在写C++代码调用.pt模型时，也要注意预处理方式要和之前保持一致，不然在C++环境下的模型预测结果会和在python环境下的结果相差很大。

![5c3b28a5c9154ad19e9e0a47c6a81bcb.png][]

[在C++环境下利用Libtorch调用pytorch模型-pudn.com][C_Libtorch_pytorch_-pudn.com]

[libtorch导入模型\_物联全栈的博客-CSDN博客][libtorch_-CSDN]

[f1e3e9cf5c884fdaa9c5178d9b33affc.png]: https://img-blog.csdnimg.cn/f1e3e9cf5c884fdaa9c5178d9b33affc.png
[5c3b28a5c9154ad19e9e0a47c6a81bcb.png]: https://img-blog.csdnimg.cn/5c3b28a5c9154ad19e9e0a47c6a81bcb.png
[C_Libtorch_pytorch_-pudn.com]: https://www.pudn.com/news/62a6b4213934cd25af818038.html
[libtorch_-CSDN]: https://blog.csdn.net/wutao22/article/details/125814844