pytorch 模型转 onnx 遇到的问题记录.

1.第一个内存溢出.

解决方法,增加系统虚拟内存.

2.导出时报错

D:\Anaconda3\lib\site-packages\torch\onnx\symbolic_helper.py:266: UserWarning: You are trying to export the model with onnx:Resize for ONNX opset version 10. This operator might cause results to not match the expected results by PyTorch.
ONNX’s Upsample/Resize operator did not match Pytorch’s Interpolation until opset 11. Attributes to determine how to transform the input were added in onnx:Resize in opset 11 to support Pytorch’s behavior (like coordinate_transformation_mode and nearest_mode).
We recommend using opset 11 and above for models using this operator.
warnings.warn("You are trying to export the model with " + onnx_op + " for ONNX opset version "
…

翻译一下

您正在尝试用导出模型onnx:Resize for ONNX opset版本10。此运算符可能导致结果与PyTorch的预期结果不匹配。
直到opset 11，ONNX的Upsample/Resize操作符才与Pytorch的插值匹配onnx:Resize in opset 11支持Pytorch的行为（比如坐标转换模式和最近的模式）。
对于使用此运算符的模型，我们建议使用opset 11及更高版本。

大概意思就是要升级一下版本. 至于是哪个版本,百度一下
You are trying to export the model with onnx:Resize for ONNX opset version 10
结果在 https://blog.csdn.net/flyfish1986/article/details/115031540 找到了答案, 感谢博主.

解决方案
默认opset version 是9，更改为11,

import torch
torch.onnx.export(model, …, opset_version=11)

后来还是提示有版本问题, 于是抱着试试看的态度我有改成了12. 结果就通了…

附加一下我的转换代码. 也是抄自别人博客的. 做了点小修改, 注意使用的时候cpu和cuda的问题. 我的cuda 内存太小于是用的cpu转的. 就是比较吃内存, 反正就运行一次. 能用就好.

    
    
    #
# demo.py
# 识别单张图片
#
import argparse
import os
import numpy as np
import time

from modeling.deeplab import *
from dataloaders import custom_transforms as tr
from PIL import Image
import torch
from torchvision import transforms
from dataloaders.utils import  *

from utils.ColorMap import colors   
 
import matplotlib
matplotlib.use('TkAgg')
import matplotlib.pyplot as plt

# from torchvision.utils import make_grid, save_image,to_image
palette = np.array(colors).reshape(-1).tolist()

def main():  
    global palette

    print("start to onnx ...") 
    parser = argparse.ArgumentParser(description="PyTorch DeeplabV3Plus Training")
    parser.add_argument('--checkpoint_file', type=str, required=True, help='image to test')
    parser.add_argument('--export_onnx_file', type=str, required=True, help='image out path')
    parser.add_argument('--backbone', type=str, default='resnet', choices=['resnet', 'xception', 'drn', 'mobilenet'],  help='backbone name (default: resnet)')
    parser.add_argument('--out_stride', type=int, default=16,  help='network output stride (default: 8)')
    parser.add_argument('--num_classes', type=int, default=2, help='crop image size')
    parser.add_argument('--sync_bn', type=bool, default=None, help='whether to use sync bn (default: auto)')
    parser.add_argument('--batch_size', type=int, default=4, help=' batch size ') 

    # parser.add_argument('--out-path', type=str, required=True, help='mask image to save')
    # parser.add_argument('--ckpt', type=str, default='deeplab-resnet.pth', help='saved model')
    # parser.add_argument('--no-cuda', action='store_true', default=False,  help='disables CUDA training')
    # parser.add_argument('--gpu-ids', type=str, default='0', help='use which gpu to train, must be a \ comma-separated list of integers only (default=0)')
    # parser.add_argument('--dataset', type=str, default='pascal', choices=['pascal', 'coco', 'cityscapes','invoice'],  help='dataset name (default: pascal)')
    # parser.add_argument('--crop-size', type=int, default=513,  help='crop image size')
    parser.add_argument('--freeze_bn', type=bool, default=True,  help='whether to freeze bn parameters (default: False)')

    args = parser.parse_args()
    # args.cuda = not args.no_cuda and torch.cuda.is_available() 

    model = DeepLab(num_classes=args.num_classes,
                    backbone=args.backbone,
                    output_stride=args.out_stride,
                    sync_bn=args.sync_bn,
                    freeze_bn=args.freeze_bn)

    ckpt = torch.load(args.checkpoint_file, map_location=lambda storage, loc: storage.cuda(0))
    # ckpt = torch.load(args.checkpoint_file, map_location='cpu')
    model.load_state_dict(ckpt['state_dict'])

    # if torch.cuda.is_available():
    #   # model = model.cuda()
    #   device = torch.device("cuda:0")
    #   model.to(device)

    model.eval()
    
    batch_size = args.batch_size #批处理大小
    input_shape = (3, 1080, 1920)   #输入数据,改成自己的输入shape
    inputx = torch.randn(args.batch_size, *input_shape)   # 生成张量
    # inputx = inputx.to('cpu')
    # inputx = inputx.cuda()
    # export_onnx_file = os.path.join(Seting.AIModelsDir_Production, "deeplabv3model.onnx")	# 目的ONNX文件名 
    torch.onnx.export(model,
                      inputx,
                      args.export_onnx_file,
                      opset_version=12,
                      do_constant_folding=True,	# 是否执行常量折叠优化
                      input_names=["input"],	# 输入名
                      output_names=["output"],	# 输出名
                      dynamic_axes={"input":{0:"batch_size"},  # 批处理变量
                                    "output":{0:"batch_size"}}
                    )

    # print("image save in image_path.")
    print("end to onnx ...")

if __name__ == "__main__":
    main()