Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

images during training are too dark #15

Open
liminn opened this issue Jun 25, 2019 · 4 comments
Open

images during training are too dark #15

liminn opened this issue Jun 25, 2019 · 4 comments

Comments

@liminn
Copy link

liminn commented Jun 25, 2019

Hi, thanks for sharing your great work!
I follow your recommendation that use the The Paris Dataset to train the network.
After training, when I check the result/train/real or result/train/cropped or result/train/recon folder, I found that all the images are too dark, such as this:
real image:
image
cropped image:
image
recon image:
recon_center_samples_epoch_199

The result of inpainting is effective, but why all the image are so dark?

ps: I do not change anything of the code except dataset/train folder.

@silence1114
Copy link

I am training now, and I also have encountered this problem. I would like to ask, after you train the model, will the result be so dark too when testing?

@liminn
Copy link
Author

liminn commented Jul 12, 2019

@silence1114 I didn't test, beacuse I thought the test during training is same as test solely. Do you have other discoveries?

@C-H-D
Copy link

C-H-D commented Aug 1, 2019

@liminn @silence1114 It's because of the normalize operationtransforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),doing unnormalization before output should solve the problem.real_cpu = real_cpu/2.0,real_cpu = real_cpu + 0.5,thenvutils.save_image(real_cpu, 'result/train/real/real_samples_epoch_%03d.png' % (epoch))

@geniusmissm
Copy link

@silence1114 ,I could you tell me how to run the procedure in windows,please!I don't quite understand .

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants