VITA-Group / DeblurGANv2

[ICCV 2019] "DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better" by Orest Kupyn, Tetiana Martyniuk, Junru Wu, Zhangyang Wang
Other
1.01k stars 265 forks source link

Can anyone reach the SSIM and PSNR results about GoPro dataset shown in paper #102

Open nests01 opened 3 years ago

nests01 commented 3 years ago

I trained the network and ran the code directly where the generator is fpn_inception and generator loss is: 0.5 ∗ Lp + 0.006 ∗ LX + 0.001 ∗ Ladv

And then I trained the network and ran the code with the same generator loss shown in paper: 0.5 ∗ Lp + 0.006 ∗ LX + 0.01 ∗ Ladv

The best results are: PSNR 27.0226; SSIM 0.8866 which are far away from the paper's results: PSNR 29.55; SSIM 0.934

Can anyone run the code on GoPro dataset and reproduce the evaluation metrics results?

lichi619 commented 3 years ago

hi,my result of repeat is PSNR 26.58,ssim 0.8968.Have you found the solution

liuwenfu1218 commented 3 years ago

same problem

peylnog commented 1 year ago

same problem