Thanks for the clarification @FrozenGene; but correct me if I’m wrong; shouldn’t HWOI be interpreted in the case of depthwise conv as HWMC which contradicts the TF layout requirements?
I found these lines: https://github.com/dmlc/tvm/blob/master/python/tvm/relay/frontend/tflite.py#L672-L674 to be in contradiction with these lines: https://github.com/dmlc/tvm/blob/master/python/tvm/relay/frontend/tflite.py#L656-L659