Researchers have developed SR-Prominence, a new dataset and protocol for evaluating artifacts in image super-resolution. This crowdsourced approach measures the "artifact prominence," which is the percentage of viewers who notice an artifact, moving beyond simple binary detection. The dataset includes 3,935 artifact masks and reveals that many previously identified artifacts are not perceived by a majority of viewers. The findings suggest that traditional metrics like SSIM and DISTS offer strong signals for prominence, while specialized artifact detectors often lack generalizability. AI
IMPACT Provides a new benchmark for evaluating the perceptual quality of AI-generated images, potentially improving future image enhancement models.
RANK_REASON Publication of a new academic paper and dataset on arXiv.
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →