Hi, thanks a lot for sharing your work! I was wondering, if PSNR is generally regarded as a metric that is unreliable when it comes to the assessment of image similarity (at least when training an SR network), why would we use it to evaluate quality of super-resolved images instead of using, say, a perceptual loss or content loss?
Hi, thanks a lot for sharing your work! I was wondering, if PSNR is generally regarded as a metric that is unreliable when it comes to the assessment of image similarity (at least when training an SR network), why would we use it to evaluate quality of super-resolved images instead of using, say, a perceptual loss or content loss?