Conv2D with Torch Creates Wrong Shape

[i418c]

While trying to move from Keras 2 to Keras 3 with PyTorch, I ran into a problem in my model where a Conv2D layer outputs a shape wider in one dimension and shorter in another. It behaves appropriately when using the Tensorflow backend.

A reproducing gist is here.

[mehtamansi29]

Hi @i418c -

Thanks for reporting the issue. Here for torch backend running on keras3 for comparing output shape need to change kernel_size and stride like this conv2d = Conv2D(output_size, kernel_size=(4,4),strides=(2, 2),padding='same',activation='relu', kernel_initializer='ones',use_bias=False).
And also input_2d need to transpose to feed input to conv2d inputs_2d = inputs_2d.transpose(1,2,3,0)

Attached gist here for the reference.

[github-actions]

This issue is stale because it has been open for 14 days with no activity. It will be closed if no further activity occurs. Thank you.