Bug in loading aligned training data

[hazirbas]

The following line will cause artifacts in the input (left edge) /output (right edge) as aligned image is upsampled with Bicubic interpolation:

pytorch-CycleGAN-and-pix2pix/data/aligned_dataset.py

Line 21 in 079da5c

AB = AB.resize((self.opt.loadSize * 2, self.opt.loadSize), Image.BICUBIC)

[junyanz]

Interesting. What is the size of your original image? What is the loadSize in your experiment? @ssnl might know the solution.

[ssnl]

Yeah, that’s possible. Maybe we can switch to first splitting and them upsampling.

…

On Fri, Mar 16, 2018 at 20:29 Jun-Yan Zhu ***@***.***> wrote: Interesting. What is the size of your original image? What is the loadSize in your experiment? @ssnl <https://github.com/ssnl> might know the solution. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#222 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AFaWZejqy8HsiF0-RHLYdwoQV9IFDNaAks5tfFjqgaJpZM4SpxFo> .

[junyanz]

That makes sense. Let me know if this commit will fix the issue. @hazirbas @ssnl