YOLO5训练好的模型部署到C++用

将YOLO5训练好的模型部署到C++需要经过以下步骤：

1. 使用PyTorch将训练好的模型导出为ONNX格式。可以使用以下代码将PyTorch模型导出为ONNX格式：

import torch
import torchvision

# Load PyTorch model
model = torchvision.models.resnet18(pretrained=False)
model.eval()

# Export the model to ONNX
input_shape = (1, 3, 224, 224)
input_names = ["input"]
output_names = ["output"]
dynamic_axes = {
    "input": {0: "batch_size"},
    "output": {0: "batch_size"},
}
torch.onnx.export(model, torch.randn(input_shape), "resnet18.onnx", input_names=input_names, output_names=output_names, dynamic_axes=dynamic_axes)

2. 安装ONNX Runtime C++库。可以从官网下载ONNX Runtime C++库，并按照指示进行安装。

3. 使用ONNX Runtime C++库加载ONNX模型并进行推理。可以使用以下代码加载ONNX模型并进行推理：

#include <iostream>
#include <string>
#include <vector>
#include <onnxruntime_cxx_api.h>

int main() {
    // Load the ONNX model
    Ort::SessionOptions session_options;
    Ort::Env env(ORT_LOGGING_LEVEL_WARNING, "test");
    Ort::Session session(env, "resnet18.onnx", session_options);

    // Get model input and output information
    Ort::AllocatorWithDefaultOptions allocator;
    Ort::TypeInfo input_type_info = session.GetInputTypeInfo(0);
    Ort::TypeInfo output_type_info = session.GetOutputTypeInfo(0);
    auto input_tensor_info = input_type_info.GetTensorTypeAndShapeInfo();
    auto output_tensor_info = output_type_info.GetTensorTypeAndShapeInfo();
    auto input_shape = input_tensor_info.GetShape();
    auto output_shape = output_tensor_info.GetShape();
    auto input_size = input_tensor_info.GetElementCount();
    auto output_size = output_tensor_info.GetElementCount();
    auto input_name = session.GetInputName(0, allocator);
    auto output_name = session.GetOutputName(0, allocator);

    // Prepare input and output buffers
    std::vector<float> input_buffer(input_size);
    std::vector<float> output_buffer(output_size);
    Ort::Value input_tensor = Ort::Value::CreateTensor<float>(allocator, input_buffer.data(), input_size, input_shape.data(), input_shape.size());
    Ort::Value output_tensor = Ort::Value::CreateTensor<float>(allocator, output_buffer.data(), output_size, output_shape.data(), output_shape.size());

    // Run inference
    Ort::RunOptions run_options;
    session.Run(run_options, input_name.c_str(), input_tensor.GetTensorMutableData<float>(), input_size * sizeof(float), output_name.c_str(), output_size * sizeof(float));
    
    // Print output
    for (int i = 0; i < output_size; i++) {
        std::cout << output_buffer[i] << std::endl;
    }

    return 0;
}

在运行前需要保证已经正确安装了ONNX Runtime C++库，并将代码中的"resnet18.onnx"替换为你自己训练好的YOLO5模型的路径。