padding type for generator

[XiaoGene]

In the paper, it is written in Section 3 that "we use mirror padding for all convolution layers". However, the code seem to use 'SAME' padding (i.e. zero padding) for the generator since the 'PADDING' field of the .yml file is specified as 'SAME'. Which type of padding did you use exactly for the pretrained models?

Thank you for your help!

[JiahuiYu]

Thanks for point out the typo. The answer is we use 'SAME' padding.

The story is: during earlier development of this work, we use mirror padding. However, we find this is almost equivalent to concatenate ones (indicating the boundaries of images) as input. See line here.

[XiaoGene]

Thanks for the answer! But why does concatenating ones as a channel indicates the boundaries of images?

[JiahuiYu]

Because the convolution automatically pad zeros as in 'SAME' mode.

[XiaoGene]

Ok, I see what you mean. Thanks!