Unable to convert Pytorch model to IR with multiple inputs

EBBIAV · ‎10-19-2021

I am currently unable to convert a Pytorch model with two inputs to IR. When I try to convert from ONNX to IR I get the following error:

[ ERROR ] Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.front.user_data_repack.UserDataRepack'>): No node with name 0

As argument for the model optimizer is use the input names 0,1 as suggested in another post:

--input "0,1"

For the conversion to ONNX I use the following arguments:

torch.onnx.export(
    model,
    (input1, input2),
    f'outputdir/model.onnx',
    input_names=["0", "1"],
    opset_version=9,
    do_constant_folding=True
)

Do you know why this happens?

Wan_Intel · ‎10-19-2021

Hi Ebbiav,

Thank you for reaching out to us!

The error you encountered: “No node name with 0” was due to Model Optimizer tried to access a node that does not exist. This could happen if you have incorrectly specified placeholder, input, or output node names.

Could you please check the input nodes of your ONNX model with Netron?

If the input nodes are 0 and 1, please share your PyTorch model, conversion scripts, and conversion command with us for further investigation.

Regards,

Wan

EBBIAV · ‎10-20-2021

Hi Wan,

Thanks for your reply!

I checked it with Netron. While my output names seem to be correctly, I cannot find the names of the input nodes (see picture below).

This is my current script for onnx conversion:

torch.onnx.export(
    model,
    (input1, input2),
    f'{OUTPUT_DIR.as_posix()}/{model.__class__.__name__}.onnx',
    input_names=["0", "1"],
    output_names=["anon_faces", "affin_matrices"],
    opset_version=9,
    do_constant_folding=True
)

I can also send other information, if that would help.

Regards,

Enrico

EBBIAV · ‎10-20-2021

The problem seems to be due to a numpy conversion of the tensor in the model. Thus, this post can be closed. Thanks for you help

Wan_Intel · ‎10-21-2021

Hi Ebbiav,

Thanks for your question.

We noticed that you would like to close this post.

If you need any additional information from Intel, please submit a new question.

Regards,

Wan