ELIQ: A Label-Free Framework for Quality Assessment of Evolving AI-Generated Images

Xinyue Li, Zhiming Xu, Zhichao Zhang, Zhaolin Cai, Sijing Wu, Xiongkuo Min, Yitong Chen, Guangtao Zhai · 2026

Read ELIQ: A Label-Free Framework for Quality Assessment of Evolving AI-Generated Images by Xinyue Li, Zhiming Xu, Zhichao Zhang, Zhaolin Cai, Sijing Wu, Xiongkuo Min, Yitong Chen, Guangtao Zhai (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03558.txt.

Description

Rating AI-generated images is a moving target: each new generator raises the quality bar and stale human labels mislead. Xinyue Li and 7 co-authors present ELIQ in a 2026 paper, a framework that assesses generated image quality with no human annotations at all. It judges both visual quality and how well the image matches its prompt.

ELIQ automatically builds positive pairs and aspect-specific negative pairs covering ordinary distortions and AI-specific defects. Those pairs teach a pretrained multimodal model, through instruction tuning, to act as a quality critic. Lightweight gated fusion and a query transformer produce a two-dimensional quality score. Across benchmarks ELIQ beats other label-free methods and transfers to user-generated content without modification.

Catalog record

CREATORXinyue Li, Zhiming Xu, Zhichao Zhang, Zhaolin Cai, Sijing Wu, Xiongkuo Min, Yitong Chen, Guangtao Zhai
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-03558

Browse the full library · About the Tower of Babel