Researchers have introduced SR-Prominence, a new dataset and protocol for evaluating image super-resolution artifacts based on perceptual impact. This approach measures "artifact prominence," defined as the proportion of viewers who notice an artifact, moving beyond simple binary detection. The dataset, comprising 3,935 artifact masks from various sources, reveals that many previously identified artifacts are not perceived by a majority of viewers. The study also found that traditional metrics like SSIM and DISTS offer strong localized prominence signals, while newer artifact detectors struggle with generalization. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Enables more accurate evaluation of image super-resolution models by focusing on perceptually relevant artifacts.
RANK_REASON Academic paper introducing a new dataset and evaluation protocol for image super-resolution artifacts. [lever_c_demoted from research: ic=1 ai=1.0]