over padding in RandomCrop when pad_if_needed

[YHRen]

in random_crop, pad_if_needed, this seems over padded the image with 2* delta_w

vision/torchvision/transforms/transforms.py

Line 448 in fe973ce

img = F.pad(img, (self.size[1] - img.size[0], 0), self.fill, self.padding_mode)

[fmassa]

This has been recently merged in #564

The logic is that if the original image is smaller than the requested crop size, there are multiple possible ways of filling it with the padding value so that the full original image is inside the crop.

By adding such padding value, we guarantee that the whole image will be inside the crop, but we give the possibility of random crops, which effectively acts as data augmentation.

I'm closing the issue, but let me know if you disagree with this reasoning.

[YHRen]

Hi Frank, I see. Thank you for the explanation. Now I understand. Bests, Ray

…

On Thu, Aug 16, 2018 at 10:13 AM, Francisco Massa ***@***.***> wrote: This has been recently merged in #564 <#564> The logic is that if the original image is smaller than the requested crop size, there are multiple possible ways of filling it with the padding value so that the full original image is inside the crop. By adding such padding value, we guarantee that the whole image will be inside the crop, but we give the possibility of random crops, which effectively acts as data augmentation. I'm closing the issue, but let me know if you disagree with this reasoning. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#576 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACOgoOjVCa6fPSoK_-d4vCLiBtWNS6JDks5uRX4jgaJpZM4V-hp8> .