They proposed an evaluation method using PIPAL, a dataset that includes various types of distortions such as GAN generation, and evaluate common metrics by Elo rating, which is often used to describe the strength of chess and other games. Evaluating the commonly used PSNR and SSIM using it, They found that they are incompatible with the perceptual results, other metrics also seem to need improvement.
TL;DR
They proposed an evaluation method using PIPAL, a dataset that includes various types of distortions such as GAN generation, and evaluate common metrics by Elo rating, which is often used to describe the strength of chess and other games. Evaluating the commonly used PSNR and SSIM using it, They found that they are incompatible with the perceptual results, other metrics also seem to need improvement.
Why it matters:
Paper URL
https://arxiv.org/abs/2007.12142
Submission Dates(yyyy/mm/dd)
Authors and institutions
Methods
Results
Comments