SR-Prominence: A Crowdsourced Protocol and Dataset Suite for Perceptually-Weighted Super-Resolution Artifact Evaluation

Researchers introduced artifact prominence, a human-centered score for how many viewers notice a marked super-resolution artifact.

They released SR-Prominence, a suite with 3,935 annotated masks spanning multiple datasets and a no-ground-truth setting.

The study found SSIM and DISTS often track localized artifact prominence better than no-reference IQA methods or specialized artifact detectors.

The work reframes super-resolution evaluation from binary defect detection to perceptual severity judged by viewers.