Pix2Pix Input Inference Image Looks Different From the Original Inference Image

[chokevin8]

When running inference on Pix2Pix, in the results folder, you get real_A, real_B, and fake_B. The problem is that all of these images look different compared to the images that I've actually inputted into the dataroot to run inference. That's really weird, since my images are all 1024 x 1024 and the --preprocess option default (resize_and_crop) shouldn't change the size of the images or translate them. All of my images are 1024 x 1024, and I did "--direction AtoB --num_test 8 --load_size 1024 --crop_size 1024" when running inference.

I can't show the images due to privacy reasons, but has anyone dealt with this weird problem during inference? I would understand if only fakeB looked different, but for some reason fakeA and realB looks different from the images that I loaded into dataroot to run inference. Thanks!

[yansonglee]

I have also encountered the same problem. Do you have a solution?@chokevin8

[arrrrr3186]

you can use '--preprocess none'