Content-variant reference IQA via knowledge distillation: a full-reference teacher transfers high-quality distribution priors to a student that needs only a non-pixel-aligned / content-variant reference, so quality scores no longer require a pixel-perfect reference image.
computer-vision deep-learning pytorch resnet knowledge-distillation ntire image-quality-assessment tid2013 full-reference-iqa csiq kadid10k mlp-mixer no-reference-iqa live-challenge koniq10k
-
Updated
Sep 6, 2026 - Python