search

ofsoundof / GRL-Image-Restoration

MIT License
379 stars 24 forks source link

two questions about the experimental results and evaluation indicators #15

Closed zxiol closed 1 year ago

zxiol commented 1 year ago

Congratulations, I have two questions about the experimental results and evaluation indicators.

  1. What do the results HQ and LQ generated by the denoising task (dn) represent? Are high-quality images and low-quality images? Does LQ represent the noisy image I input?
  2. Which two images are used to calculate the evaluation indicators PSNR and SSIM? Because I seem to have only inputted noisy images without corresponding reference images. What does the 'y' in 'psnr_y' represent? Thank you very much for your reply Snipaste_2023-04-17_20-12-33 Snipaste_2023-04-17_20-13-14
ofsoundof commented 1 year ago

Thanks a lot for the question! The following are my answers.

  1. HQ: the restored high-quality images. For image denoising, this is the denoised image.
  2. LQ: this is the low-quality image. For image denoising, it is the noisy input.
  3. The ground-truth image and the restored images (HQ) are used to calculate PSNR and SSIM. Actually, for Gaussian image denoising, the input of the dataloader is the ground-truth images and the noisy images are generated on the fly.
  4. Y means the Y channel. So psnr_y is the PSNR calculated on the Y channel. And psnr without y suffix is the PSNR calculated on the RGB channels.