onnx模型转换成paddle，发生报错：AssertionError: only dilations = 0 is supported以及报错Exception: convert failed node:_stem_pool_MaxPool_output_0, op_type is MaxPool

[17858569330]

感谢您参与 X2Paddle 社区！ 问题模版为了 X2Paddle 能更好的迭代，例如新功能发布、 RoadMaps 和错误跟踪. 😸

问题描述

D:\DESKTOP\CVNet-main
x2paddle --framework=onnx --model=CVNet_Rerank.onnx --save_dir=pd_model
model ir_version: 6, op version: 11
Shape inferencing ...
[WARNING] Incomplete symbolic shape inference
Shape inferenced.
Now, onnx2paddle support convert onnx model opset_verison [7, 8, 9, 10, 11, 12, 13, 14, 15], opset_verison of your onnx model is 11.
Total nodes: 1806
Nodes converting ...
Converting node 1050 ... Traceback (most recent call last):
File "C:\Users\msipc.conda\envs\sam\Lib\site-packages\x2paddle\op_mapper\onnx2paddle\opset_legacy.py", line 111, in run_mapping
res = func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\msipc.conda\envs\sam\Lib\site-packages\x2paddle\op_mapper\onnx2paddle\opset_legacy.py", line 1977, in MaxPool
assert node.get_attr(
^^^^^^^^^^^^^^
AssertionError: only dilations = 0 is supported

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "", line 198, in _run_module_as_main
File "", line 88, in run_code
File "C:\Users\msipc.conda\envs\sam\Scripts\x2paddle.exe_main.py", line 7, in
File "C:\Users\msipc.conda\envs\sam\Lib\site-packages\x2paddle\convert.py", line 501, in main
onnx2paddle(args.model,
File "C:\Users\msipc.conda\envs\sam\Lib\site-packages\x2paddle\convert.py", line 316, in onnx2paddle
mapper = ONNXOpMapper(model)
^^^^^^^^^^^^^^^^^^^
File "C:\Users\msipc.conda\envs\sam\Lib\site-packages\x2paddle\op_mapper\onnx2paddle\onnx_op_mapper.py", line 53, in init
func(node)
File "C:\Users\msipc.conda\envs\sam\Lib\site-packages\x2paddle\op_mapper\onnx2paddle\opset_legacy.py", line 113, in run_mapping
raise Exception("convert failed node:{}, op_type is {}".format(
Exception: convert failed node:_stem_pool_MaxPool_output_0, op_type is MaxPool

具体信息

需要将CVNet的模型权重文件转换成paddle。

• 转换模型后用处

  • 使用 Paddle 框架/ PaddleInference 推理预测
  • 转换预训练参数，再使用 Paddle 进行模型开发

• 模型来源
  https://github.com/sungonce/cvnet

• 联系方式
  [email protected]
  [email protected]
  17858560330