How Accurate are Video Quality Models for Diffusion-Based Video Super-Resolution?

A new study evaluated six video upscaling methods on both compressed and uncompressed low-resolution videos.

CNN-based full-reference metrics tracked human ratings better than other automated models, but still made notable mistakes.

Most no-reference and full-reference quality models were not reliable enough to replace subjective testing for diffusion-based video super-resolution.

The authors released the videos, human ratings, and model scores as open benchmark data.