Hi,

I was just wondering whether it is possible/planned to add ONNX support for AdaptivePooling, because that enables using the same network with different resolutions (check https://github.com/fastai/fastai/blob/84d8ad253e1d3f5ecf1e1dcb07076e61f3f408b7/fastai/conv_learner.py#L40) for example.
When I wanted to export this it failed with

RuntimeError: ONNX export failed: Couldn't export Python operator AdaptiveMaxPool2d

and I had to manually replace the AdaptivePoolings with normal ones, thus it now works with fixed sized inputs.

Cheers, Johannes

Edit: So in this case its possible to use GlobalPooling but there are still use cases for AdaptivePooling, where ONNX suppport would be very nice