why there is a pooling layer with pool_size=(1,1) before the upsamping layer?

[ssiaw12345]

As the title said, there is a pooling layer with pool_size=(1,1) before the upsamping layer which I think don't make sense, such as face.model.build_model line 65

[somewacko]

Hmm I think that when I first implemented it I had it in my head that deconv was basically reverse conv/pooling, which is why that's there. It's mostly harmless, although I wonder if it's unnecessarily eating up GPU memory since it would just make a copy of its input.

[ssiaw12345]

Thank you for your prompt reply~

[somewacko]

Yeah thanks for pointing this out ! -- I just removed it and am able to train with larger batch sizes now, so I think it was just eating memory. My bad~