Does onnx2torch support dynamic input shape?

[liamsun2019]

Hi author,
I am curious if onnx2torch supports dynamic input shape. I try to convert an onnx whose input shape is dynamic, i.e, [batch, 10, 20, 30]. The conversion looks normal where no errors rise up. But the inference based on the converted model fails with following errors:

RuntimeError: Given groups=1, weight of size [xx, yy, zz], expected input[aa, bb, cc] to have XX channels, but got YY channels instead

The code snippet:
model = onnx2torch.convert(onnx_model)
model(*dummy_input)

Any comments? Thanks.

[senysenyseny16]

hi @liamsun2019,

In general case - yes. Could you provide model or minimal example to reproduce error?