TorchVision.utils.save_images() saves images with additional offsets

Saves input images of dimension 128x128 to 132x132. The boundary pixels are relevant in my case and I resue them for other network operations. Resizing them seems like not a good idea. Any workaround?

You can remove the padding that is aded to the images by default.

I set padding=0.
I want to save image with size 256 * 256.
But I get 257*257 with additional 1 resolution offsets

Save image I guess is just to save your batch in a grid way. You can just use the torchvision toPilImage and save your images one by one