onnx转ncnn的问题

[queenmary55]

pytorch模型转onnx，使用rvm作者给的代码，没报错，但是有警告:
onnx_helper.py:20: TracerWarning: Converting a tensor to a Pyth on number might cause the trace to be incorrect. We can't record the data flow of Python values, so this value wil l be treated as a constant in the future. This means that the trace might not generalize to other inputs! return F.interpolate(x, scale_factor=scale_factor.item(),

使用该onnx模型，进行简化，命令如下：
python -m onnxsim model.onnx model_sim.onnx --dynamic-input-shape --input-shape src:1,3,1280,720 r1i:1,16,256,144 r2i:1,20,128,72 r3i:1,40,64,36 r4i:1,64,32,18 downsample_ratio:1
报错如下
onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Non-zero status code returned while running Expand node. Name:'Expand_174' Status Message: invalid expand shape

请大神赐教。我用netron查看了onnx模型的网络图，找到了Expand_174节点，该节点是接收一个shape，然后对r4i进行操作（我去onnx的github上查看，说expand操作是广播）onnx中expand操作的解释
但是我仔细查看并打印了decoder.py里面相关位置的tensor形状，没有自动广播的情况

[FeiGeChuanShu]

这个跟onnx的版本也有关系，会给你搞出很多莫名其妙的op，要不你试一下pnnx？ 可以直接torchscript转ncnn，不会涉及onnx这个中间商。 https://github.com/Tencent/ncnn/tree/master/tools/pnnx

[FeiGeChuanShu]

@queenmary55

[queenmary55]

@FeiGeChuanShu 感谢回复！我去试试pnnx

[queenmary55]

@FeiGeChuanShu 请教一下，使用pnnx进行转换的过程中，是否需要对原网络结构的代码进行修改，您在转换的时候有什么注意事项么，我尝试从torchscript用pnnx转ncnn报错如下
。。。。。。。。
############# pass_level2
############# pass_level3
############# pass_level4
############# pass_level5
make_slice_expression input 173
make_slice_expression end 174
terminate called after throwing an instance of 'std::out_of_range'
what(): map::at
Aborted (core dumped)

[FengMu1995]

@queenmary55 问下你上面的问题了吗

[queenmary55]

@FengMu1995 没有解决。我感觉是level1的时候，就有不支持的算子，在encoder部分的featuremap形状就不太对，最后在decoder部分报错了，out of range。不知道怎么排查错误

[queenmary55]

@FengMu1995 方便留个邮箱联系下吗

[li-wenquan]

@queenmary55 请问您后来如何转换成功？是否愿意分享一下？谢谢。即使我使用了onnxsim简化，最后再转换为ncnn时报以下错误(似乎onnxsim没有作用？)：
Unknown data type 0
Unsupported Resize scales and sizes are all empty!
Unsupported split axis !
Shape not supported yet!
Expand not supported yet!
Shape not supported yet!
Unknown data type 0
Shape not supported yet!
Expand not supported yet!
Shape not supported yet!
Unknown data type 0
Shape not supported yet!
Expand not supported yet!
Shape not supported yet!
Unknown data type 0
Shape not supported yet!
Expand not supported yet!
Shape not supported yet!
Unknown data type 0
Unsupported split axis !
Shape not supported yet!
Shape not supported yet!
Unknown data type 0
Unsupported Resize scales and sizes are all empty!
Shape not supported yet!
Shape not supported yet!
Unknown data type 0
Unsupported Resize scales and sizes are all empty!

[rose-jinyang]

Hi @FeiGeChuanShu
How are you?
Thanks for contributing to this project.
I am going to use your NCNN model in Visual Studio C++ on Windows 10.
But it is impossible to run your NCNN model.
First, I see a message which shows that your ncnn model is too old.
Could u share the method for converting the original rvm model to ncnn model?
Thanks