自己训练的yolov5模型转ONNX转trt（记录）

1.模型导出ONNX

在官方提供的代码中改了两句，输入和模型都加载CPU上，否则报错模型是tensor输入不是或者输入是tensor模型不是。

opset_version = 10这里需要注意，设置太大或太小都转换不成功，第一次设置11的时候转trt不成功，改为10的话就没问题了。

"""Exports a pytorch *.pt model to *.onnx format

Usage:
    import torch
    $ export PYTHONPATH="$PWD" && python models/onnx_export.py --weights ./weights/yolov5s.pt --img 640 --batch 1
"""

import argparse

import onnx

from models.common import *

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--weights', type=str, default='weights/best_can.pt', help='weights path')
    parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='image size')
    parser.add_argument('--batch-size', type=int, default=1, help='batch size')
    parser.add_argument('--device', default='cpu', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
    opt = parser.parse_args()
    print(opt)

    device = torch_utils.select_device(opt.device)
    # Parameters
    f = opt.weights.replace('.pt', '.onnx')  # onnx filename
    # img = torch.zeros((opt.batch_size, 3, *opt.img_size))  # image size, (1, 3, 320, 192) iDetection
    img = torch.zeros((1, 3, opt.img_size[0], opt.img_size[1]), device=device)

    # Load pytorch model
    google_utils.attempt_download(opt.weights)
    model = torch.load(opt.weights, map_location=device)['model']
    model.eval()
    model.fuse()

    # Export to onnx
    model.model[-1].export = True  # set Detect() layer export=True
    _ = model(img)  # dry run
    torch.onnx.export(model, img, f, verbose=False, opset_version=10, input_names=['images'],
                      output_names=['output'])  # output_names=['classes', 'boxes']

    # Check onnx model
    model = onnx.load(f)  # load onnx model
    onnx.checker.check_model(model)  # check onnx model
    print(onnx.helper.printable_graph(model.graph))  # print a human readable representation of the graph
    print('Export complete. ONNX model saved to %s\nView with https://github.com/lutzroeder/netron' % f)

2.为了减少后面的报错，把ONNX转为ONNX_SIM

pip install onnx-simplifier

python -m onnxsim best.onnx best_sim.onnx

3.对于转换完的模型可以看一下网络结构，或者用onnx_sim推理一下查看一下推理效果