search

Huang-ShiRui / Semi-UIR

Official code for "Contrastive Semi-supervised Learning for Underwater Image Restoration via Reliable Bank" (2023 CVPR)
113 stars 10 forks source link

Running with your pretrained model to get the reported PSNR #13

Closed weilegexiang closed 1 year ago

weilegexiang commented 1 year ago

Hi, I greatly admire your work~ I attempted to reproduce the PSNR results reported in your Tab. 1 of the paper, but I only get 22.9 on testR (the reported one is 24.6), and 22.3 on testS (23.4), the SSIM is identical to your paper. Is there something wrong with the provided pretrained model? or if you can share me with the test code for reproducing the PSNR.

Thank you !

weilegexiang commented 1 year ago

No question~ Thank you ~

zxiol commented 1 year ago

Hello, may I ask which specific images you used to test the model? It seems that the author did not provide a specific partition of the dataset he used. Thank you!

weilegexiang commented 1 year ago

Hello, may I ask which specific images you used to test the model? It seems that the author did not provide a specific partition of the dataset he used. Thank you!

Hi, they may reported the PSNR-y other than the PSNR.

zxiol commented 1 year ago

Thank you very much. The author mentioned in the paper that 800 pairs of images were used from a synthetic dataset, but I am not sure how 800 pairs of images were selected from ten types of water quality. Do you know?

weilegexiang commented 1 year ago

Thank you very much. The author mentioned in the paper that 800 pairs of images were used from a synthetic dataset, but I am not sure how 800 pairs of images were selected from ten types of water quality. Do you know?

I have no their data for training and testing, but as I known, they ever released their whole datasets, but now, it's gone.

zxiol commented 1 year ago

So it seems difficult to reproduce the results of the paper. Is the pre trained model provided by the author the final model of the paper?https://github.com/Huang-ShiRui/Semi-UIR/tree/14eb75d78a405cdc1f259a5ccc84dc90a63e8c13/pretrained

weilegexiang commented 1 year ago

So it seems difficult to reproduce the results of the paper. Is the pre trained model provided by the author the final model of the paper?https://github.com/Huang-ShiRui/Semi-UIR/tree/14eb75d78a405cdc1f259a5ccc84dc90a63e8c13/pretrained

I think the provided model can get the results reported in the paper, I tested several real images, and it seems the model works.

zxiol commented 1 year ago

Thank you for your answer. Good luck