[BUGs] Backward error for non-square images

[Younger330]

Hello,

I am experiencing an issue with backward errors after using downsampling on feature maps of dimension (Batch_size, 256, 360, 480). After performing the backward operation, an error is returned even though the dimensions are verified to be correct.

output = right_pool.backward(input, grad_output)[0]
RuntimeError: The size of tensor a (480) must match the size of tensor b (360) at non-singleton dimension 2

Interestingly, when I replace the C++ file with https://github.com/princeton-vl/CornerNet-Lite/tree/master/core/models/py_utils/_cpools , the error no longer occurs.

Could you please look into this issue?

Thank you for your assistance.