脚本之家,脚本语言编程技术及教程分享平台!
分类导航

Python|VBS|Ruby|Lua|perl|VBA|Golang|PowerShell|Erlang|autoit|Dos|bat|

服务器之家 - 脚本之家 - Python - Pytorch模型转onnx模型实例

Pytorch模型转onnx模型实例

2020-04-21 09:47joey.lei Python

今天小编就为大家分享一篇Pytorch模型转onnx模型实例,具有很好的参考价值,希望对大家有所帮助。一起跟随小编过来看看吧

如下所示:

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
import io
import torch
import torch.onnx
from models.C3AEModel import PlainC3AENetCBAM
 
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
def test():
  model = PlainC3AENetCBAM()
 
  pthfile = r'/home/joy/Projects/models/emotion/PlainC3AENet.pth'
  loaded_model = torch.load(pthfile, map_location='cpu')
  # try:
  #   loaded_model.eval()
  # except AttributeError as error:
  #   print(error)
 
  model.load_state_dict(loaded_model['state_dict'])
  # model = model.to(device)
 
  #data type nchw
  dummy_input1 = torch.randn(1, 3, 64, 64)
  # dummy_input2 = torch.randn(1, 3, 64, 64)
  # dummy_input3 = torch.randn(1, 3, 64, 64)
  input_names = [ "actual_input_1"]
  output_names = [ "output1" ]
  # torch.onnx.export(model, (dummy_input1, dummy_input2, dummy_input3), "C3AE.onnx", verbose=True, input_names=input_names, output_names=output_names)
  torch.onnx.export(model, dummy_input1, "C3AE_emotion.onnx", verbose=True, input_names=input_names, output_names=output_names)
 
if __name__ == "__main__":
 test()

直接将PlainC3AENetCBAM替换成需要转换的模型,然后修改pthfile,输入和onnx模型名字然后执行即可。

注意:上面代码中注释的dummy_input2,dummy_input3,torch.onnx.export对应的是多个输入的例子。

在转换过程中遇到的问题汇总

RuntimeError: Failed to export an ONNX attribute, since it's not constant, please try to make things (e.g., kernel size) static if possible

在转换过程中遇到RuntimeError: Failed to export an ONNX attribute, since it's not constant, please try to make things (e.g., kernel size) static if possible的错误。

根据报的错误日志信息打开/home/joy/.tensorflow/venv/lib/python3.6/site-packages/torch/onnx/symbolic_helper.py,在相应位置添加print之后,可以定位到具体哪个op出问题。

例如:

在相应位置添加

?
1
print(v.node())

输出信息如下:

?
1
%124 : Long() = onnx::Gather[axis=0](%122, %121), scope: PlainC3AENetCBAM/Bottleneck[cbam]/CBAM[cbam]/ChannelGate[ChannelGate] # /home/joy/Projects/models/emotion/WhatsTheemotion/models/cbam.py:46:0

原因是pytorch中的tensor.size(1)方式onnx识别不了,需要修改成常量。

以上这篇Pytorch模型转onnx模型实例就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。

原文链接:https://blog.csdn.net/lei19880402/article/details/103721362

延伸 · 阅读

精彩推荐