search

haoyuc / LWay

[CVPR 2024] Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning
https://arxiv.org/abs/2403.02601
20 stars 0 forks source link

Question about quantitative comparison results #5

Open ZahraFan opened 1 month ago

ZahraFan commented 1 month ago

Hi! The work is inspiring!

However, I have some doubts while reading the paper. The metric performance values for several baseline methods in Table 1 do not seem to match what I have seen in related papers. For example, StableSR tested on RealSR gives PSNR\SSIM\LPIPS values of 24.65\0.7080\0.3002, which appear worse than the data shown in your paper. If I have missed some details, please point them out.

Could you provide some information about the testing process?

haoyuc commented 1 month ago

Thank you for your interest in our work and for your careful observation.

You're correct. This discrepancy is due to our different testing methodology. We independently preprocessed all test images by cropping them to ensure that the corresponding high-resolution (HR) size was 512x512 pixels before conducting our tests. This cropping step likely differs from the original approach used in the StableSR paper. As a result, our reported metrics don't match those in the original paper.