search

sswuai / PanoGAN

Other
38 stars 7 forks source link

Pytorch version #1

Open qianmingduowan opened 2 years ago

qianmingduowan commented 2 years ago

Hi When I use your pretrained model to predict image on op dataset, I only get average_psnr and ssim : 13.055474564891604 , 0.3293401456411093 I calculate psnr and ssim by skimage.metrics package. So, I think there is no wrong with calculating metrics.

I want know which pytorch version you used? I can not get a good result with pytorch 1.12.0 and 1.8.2. Wish for your reply.

Ha0Tang commented 2 years ago

we use this file https://github.com/sswuai/PanoGAN/blob/master/evaluation/compute_ssim_psnr_sharpness.lua to calculate psnr and ssim

qianmingduowan commented 2 years ago

we use this file https://github.com/sswuai/PanoGAN/blob/master/evaluation/compute_ssim_psnr_sharpness.lua to calculate psnr and ssim

Thanks for your reply. But we all know that psnr is easy to be calculated without any hyperparameters. I tried three different PNSR calculate codes, and I got the same PSNR value on the op dataset: 13.055474564891604 I think the different versions of PyTorch may cause it. When I use a higher PyTorch version to load your pretrained model, it may cause some unimagine results. So I still want to know your PyTorch version. Or there are some mistakes in your released pre-trained weight?

sswuai commented 1 year ago

Hi, mingqian,

Thank you for your attention to our work.

Our paper's experiments were performed on our old server, whose hardware and software (including cuda version, python version and pytorch version) have been updated. I need two or three days to check the log to retrieve the detailed experimental environment information.

I will try my best to provide you asap with the necessary information for result reproduction.

Best regards,

Sognsong

On Fri, Jul 8, 2022 at 12:49 AM mingqian @.***> wrote:

we use this file https://github.com/sswuai/PanoGAN/blob/master/evaluation/compute_ssim_psnr_sharpness.lua to calculate psnr and ssim

Thanks for your reply. But we all know that psnr is easy to be calculated without any hyperparameters. I tried three different PNSR calculate codes, and I got the same PSNR value on the op dataset: 13.055474564891604 I think the different versions of PyTorch may cause it. When I use a higher PyTorch version to load your pretrained model, it may cause some unimagine results. So I still want to know your PyTorch version. Or there are some mistakes in your released pre-trained weight?

— Reply to this email directly, view it on GitHub https://github.com/sswuai/PanoGAN/issues/1#issuecomment-1177921451, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADD5CXJQLBRJP3BF647CKJ3VS4DDFANCNFSM524G35MA . You are receiving this because you are subscribed to this thread.Message ID: @.***>