Failed to convert ONNX neural network

I’m trying to convert a neural network. In the conversion script I issue a command like (convert_onnx is properly defined):
–onnx-model ./network.onnx
–net-output ${NAME}.json
–data-output ${NAME}.data

The conversion fails with the following output:
I Current ONNX Model use ir_version 4 opset_version 9
D This op Conv of Conv_1 not able get out tensor Conv_1:out0 shape
I build output layer attach_Sigmoid_118:out0
I build output layer attach_Sigmoid_120:out0
I Try match Sigmoid_120:out0
W Not match tensor Sigmoid_120:out0
Traceback (most recent call last):
File “”, line 25, in
File “”, line 20, in main
File “acuitylib/app/importer/”, line 44, in run
File “acuitylib/converter/”, line 1005, in match_paragraph_and_param
File “acuitylib/converter/”, line 918, in _onnx_push_ready_tensor
TypeError: ‘NoneType’ object is not iterable
[2725] Failed to execute script convertonnx

The network architecture is as follows:

Is there a way to figure out what’s wrong?

Is there a way to obtain source code for these conversion scripts so that i could try to figure out myself. Otherwise i’m stuck with no way out. Without solving the problem this is just a piece of useless hardware for me.

Hello, Khadas is on holiday, more help will be available when they return.

Bumping this up. Can please somebody look at the issue?

@Frank Please help on this

@sermus I am sorry that the source code for these conversion scripts wasn’t open source . We don’t have the source code for this tool.And can you tell me which platform you used to train this model, Tensorflow ? caffe or other ?

Error message: None Type cannot be iterated. I think it’s related to the platform of model training. This NPU model and tools does not support the newer platform version very well at present, maybe there will be problems

@Frank Can you please elaborate why this is important? The model was trained in PyTorch and then converted into ONNX. This is a valid ONNX model because checker tools don’t find any errors. Even if it’s not valid from converter’s stand point i need to know where this invalidity lies in. Now i get just stack trace for the binaries i don’t have source code for. This doesn’t provide any clue where to start analysis from.

@sermus First, the whole transformation environment needs to rely on some tools of tensorflow. Even if you are not using tensorflow model, other models are also needed. Secondly, this conversion tool and this version of SDK only support tf1.10. Installing other versions of tensorflow will lead to similar bug of data type recognition error during conversion.

@sermus So,I asked you about the version, which is very important. Because the conversion environment is not installed correctly, it cannot be converted normally