search

cschenxiang / DRSformer

Learning A Sparse Transformer Network for Effective Image Deraining (CVPR 2023)
254 stars 14 forks source link

Bad resutls on Rain100L #15

Closed mikasaq closed 10 months ago

mikasaq commented 10 months ago

Hi, I'm retraining DRSformer on rain100L using Deraining.yml and DRSformer_200L+SPA_arch.py. But the avarage psnr and ssim on y channel are only 40.05 and 0.986, respectively. Is there something I did not notice? The loss and psnr during training are shown below. image

cschenxiang commented 10 months ago

Hi, I'm retraining DRSformer on rain100L using Deraining.yml and DRSformer_200L+SPA_arch.py. But the avarage psnr and ssim on y channel are only 40.05 and 0.986, respectively. Is there something I did not notice? The loss and psnr during training are shown below. image

Our experiment was conducted on the Rain200L dataset. Check the settings for multi GPU training and batch size, and ensure that the code for calculating PSNR is consistent.

mikasaq commented 10 months ago

Thank you for your reminder.