output_feature has no shape when output has len(shape)==2

[yuvdar]

❓Question

in _make_coreml_output_features in converter.py you don't allow shape of 2 (B, C), why is it? this is quite common in classification networks (such as mobileNet).

[bhushan23]

@yuvdar we do handle output of rank 2 convert adds output feature with shape none is that case. Have you encountered any case we don't handle this? please share an example if that's the case

[yuvdar]

@bhushan23 - Thanks for the reply. This is indeed the case. My question is why not give out the correct output shape? None is not informative.

[bhushan23]

@yuvdar when you set None, output shape will be inferred

[yuvdar]

@bhushan23 The problem is, there is no way of knowing the shape before inference. The metadata doesn't provide it, while in other cases it does.