Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

TL;DR

ELIQ tackles the problem of quality assessment for evolving AI-generated images where traditional MOS-based labels drift with generative models. It introduces a label-free framework that automatically constructs high-quality positives and aspect-specific negatives to supervise a two-dimensional quality space: visual quality and prompt-image alignment, via quality-aware instruction tuning, gated fusion, and a lightweight Quality Query Transformer. The method achieves state-of-the-art performance among label-free approaches and demonstrates strong transfer to user-generated content without architectural changes, reducing reliance on costly annotations. By leveraging multimodal language models and carefully designed degradations, ELIQ provides scalable, continually updatable IQA suitable for rapidly advancing AIGC ecosystems. The work highlights a practical path toward sustainable quality evaluation as generative models evolve, with broad implications for benchmarking, development, and deployment of AI-generated imagery.

Abstract

Generative text-to-image models are advancing at an unprecedented pace, continuously shifting the perceptual quality ceiling and rendering previously collected labels unreliable for newer generations. To address this, we present ELIQ, a Label-free Framework for Quality Assessment of Evolving AI-generated Images. Specifically, ELIQ focuses on visual quality and prompt-image alignment, automatically constructs positive and aspect-specific negative pairs to cover both conventional distortions and AIGC-specific distortion modes, enabling transferable supervision without human annotations. Building on these pairs, ELIQ adapts a pre-trained multimodal model into a quality-aware critic via instruction tuning and predicts two-dimensional quality using lightweight gated fusion and a Quality Query Transformer. Experiments across multiple benchmarks demonstrate that ELIQ consistently outperforms existing label-free methods, generalizes from AI-generated content (AIGC) to user-generated content (UGC) scenarios without modification, and paves the way for scalable and label-free quality assessment under continuously evolving generative models. The code will be released upon publication.

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

TL;DR

ELIQ tackles the problem of quality assessment for evolving AI-generated images where traditional MOS-based labels drift with generative models. It introduces a label-free framework that automatically constructs high-quality positives and aspect-specific negatives to supervise a two-dimensional quality space: visual quality and prompt-image alignment, via quality-aware instruction tuning, gated fusion, and a lightweight Quality Query Transformer. The method achieves state-of-the-art performance among label-free approaches and demonstrates strong transfer to user-generated content without architectural changes, reducing reliance on costly annotations. By leveraging multimodal language models and carefully designed degradations, ELIQ provides scalable, continually updatable IQA suitable for rapidly advancing AIGC ecosystems. The work highlights a practical path toward sustainable quality evaluation as generative models evolve, with broad implications for benchmarking, development, and deployment of AI-generated imagery.

Abstract

Generative text-to-image models are advancing at an unprecedented pace, continuously shifting the perceptual quality ceiling and rendering previously collected labels unreliable for newer generations. To address this, we present ELIQ, a Label-free Framework for Quality Assessment of Evolving AI-generated Images. Specifically, ELIQ focuses on visual quality and prompt-image alignment, automatically constructs positive and aspect-specific negative pairs to cover both conventional distortions and AIGC-specific distortion modes, enabling transferable supervision without human annotations. Building on these pairs, ELIQ adapts a pre-trained multimodal model into a quality-aware critic via instruction tuning and predicts two-dimensional quality using lightweight gated fusion and a Quality Query Transformer. Experiments across multiple benchmarks demonstrate that ELIQ consistently outperforms existing label-free methods, generalizes from AI-generated content (AIGC) to user-generated content (UGC) scenarios without modification, and paves the way for scalable and label-free quality assessment under continuously evolving generative models. The code will be released upon publication.
Paper Structure (33 sections, 21 equations, 3 figures, 5 tables)

This paper contains 33 sections, 21 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. AIGC Generation Models
  4. Image Quality Assessment
  5. Multi-modal Model-based IQA
  6. Label-Free IQA Methods
  7. Proposed Method
  8. Framework Overview
  9. Label-Free Supervision Construction
  10. Prompt Selection
  11. Positive Sample Generation
  12. Generation of Negative Samples
  13. Model Training
  14. Quality-aware Instruction Tuning
  15. Aspect-specific instructions and negatives.
  16. ...and 18 more sections

Figures (3)

  • Figure 1: The rapid evolution of generative models shifts MOS distributions, making annotations increasingly inconsistent. ELIQ replaces absolute MOS labels with automatically constructed supervision, enabling scalable quality assessment for evolving AIGC.
  • Figure 2: Overview of label-free positive and aspect-specific negative sample construction. High-quality images are generated from curated prompts using multiple T2I models, while negative samples are created by simulating technical, aesthetic, and alignment degradations, including both conventional distortions and AI-specific generation artifacts.
  • Figure 3: Overview of the proposed label-free visual and alignment quality scoring framework.