search

NVIDIA / pix2pixHD

Synthesizing and manipulating 2048x1024 images with conditional GANs
https://tcwang0509.github.io/pix2pixHD/
Other
6.67k stars 1.39k forks source link

Bad results when testing #162

Open ShaniGam opened 5 years ago

ShaniGam commented 5 years ago

I train the model on 2975 images from the cityscapes dataset with the following command: python train.py --name label2city_512p_feat_cityscapes --instance_feat The outputs I get during the training seem fine: epoch054_160500_synthesized_image epoch055_161000_synthesized_image

But when I test the model (on images from the training set) I get:

aachen_000004_000019_gtFine_labelIds_synthesized_image aachen_000010_000019_gtFine_labelIds_synthesized_image

Does anyone has an idea of why it happens?

cszer commented 4 years ago

overfitting

wang-zm18 commented 4 years ago

Did you solve the problem? I also meet with this problem during testing.

yxwang1794 commented 3 years ago

can we add a validation set to prevent overfitting in pix2pixHD model? I haven't seen anyone done this before.

edd2110-jac commented 3 years ago

overfitting

I don't think it's because overfitting. He says that he is "Testing the model with images from the training set". If it is overfitting. It shouldn't work with any other images but with images the model was trained with should work.

I'm having this same issue, with no good outcome yet.

songyn95 commented 2 years ago

I'm having this same issue, How did you fix it? @ShaniGam @wang-zm18 @cszer @yxwang1794 @edd2110-jac

PradeepYLN commented 1 month ago

@ShaniGam @wang-zm18 @cszer @yxwang1794 @edd2110-jac @songyn95 Has anyone added the validation step to check for overfitting?