Add AdaptiveAvgPool2d and AdaptiveMaxPool2d to ONNX.symbolic

[ksemianov]

Add AdaptiveAvgPool2d and AdaptiveMaxPool2d to ONNX.symbolic
Due to limitations in ONNX only output_size=1 is supported.
AdaptiveAvgPool2d -> GlobalAveragePool
AdaptiveMaxPool2d -> GlobalMaxPool
Fixes #5310

[facebook-github-bot]

ezyang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[weihua04]

@ksemianov , my pytorch model has a nn.AdaptiveAvgPool2d((3, 3)), and i can't export onnx model with error " only output_size=1 is supported",do u have any good idea to solve the problem?
thank u very much

[ksemianov]

@weihua04 GlobalAveragePool in ONNX doesn't have output_size attribute. It implicitly assumes that output_size=1. You might want to workaround with nn.AvgPool2d(input_size - (output_size - 1) * (input_size // output_size), stride=input_size // output_size). If input_size is divisible by output_size, it will give exactly the same result. Otherwise, there'll be some prediction error.