search

yunjey / stargan

StarGAN - Official PyTorch Implementation (CVPR 2018)
MIT License
5.21k stars 967 forks source link

Low quality results by using pretrained model 256 on celebA dataset #113

Open Yummy2016 opened 4 years ago

Yummy2016 commented 4 years ago

I'm using 256*256 model to test some pictures in celebA dataset, arguments are set as:

python main.py --mode test --dataset CelebA --image_size 256 --c_dim 5 \ --selected_attrs Black_Hair Blond_Hair Brown_Hair Male Young \ --model_save_dir='stargan_celeba_256/models' \ --result_dir='stargan_celeba_256/results'

However, the results' quality are far from the results posted on the paper. The faces are blurring and not changed that much on expressions or other selected attributes.

16-images

Are there any places that I need to tune to get the results with the same quality that the paper posted?

Thank you!

07hyx06 commented 4 years ago

I got the same bad result.....Do you solve it?

Yummy2016 commented 4 years ago

no.....

On Wed, Jan 1, 2020 at 6:24 PM Yuxuan Han notifications@github.com wrote:

I got the same bad result.....Do you solve it?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/yunjey/stargan/issues/113?email_source=notifications&email_token=AFJLPEST2KSZ6DQ57AMICBDQ3RVPRA5CNFSM4JWDH7JKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEH5CCMA#issuecomment-570040624, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFJLPEUY66I3BBRAKXCRW2DQ3RVPRANCNFSM4JWDH7JA .

07hyx06 commented 4 years ago

no..... … On Wed, Jan 1, 2020 at 6:24 PM Yuxuan Han @.***> wrote: I got the same bad result.....Do you solve it? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#113?email_source=notifications&email_token=AFJLPEST2KSZ6DQ57AMICBDQ3RVPRA5CNFSM4JWDH7JKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEH5CCMA#issuecomment-570040624>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFJLPEUY66I3BBRAKXCRW2DQ3RVPRANCNFSM4JWDH7JA .

I tried 128*128 pretrained model, its behavior is the same as paper.....

Yummy2016 commented 4 years ago

yes, 128*128 is better than 256*256. not sure what's the reason...

fangxuehouwuming commented 9 months ago

Do you solve it?

yyyoite commented 6 months ago

Me too