IQA-T1: Making LLMs Gather Evidence Before Scoring Images
量子位 · wechat · 2026-07-20
This post introduces IQA-T1, a multimodal image quality assessment framework designed to stop LLMs from scoring images based on mere "feel." Instead, the model first calls specialized visual tools to generate evidence, then reasons based on that evidence.
The article points out that current MLLMs are easily swayed by semantic biases during IQA: they excel at recognizing "what this is" but struggle to capture low-level degradations like noise, blur, or compression artifacts. To fix this, IQA-T1 features 15 visual evidence tools covering attributes like noise residuals, Fourier spectra, and gradient consistency. It uses a two-stage training approach (SFT + GRPO) to teach the model "how to use tools" and "when to use them."
The team also built the Q-Tool dataset containing 11k reasoning chains, achieving state-of-the-art overall results across 7 IQA benchmarks with average PLCC/SRCC scores of 0.795/0.784. The article emphasizes that beyond delivering more accurate scores, this method makes the reasoning process explainable and traceable.
More from Research
- Why a 1GW Chinese AI data center may be plausible after all — teortaxesTex · 2026-07-22
- Chinese AI labs are now treating distillation obfuscation as the top research topic — pmddomingos · 2026-07-22
- Structural ensembles beat single predictions in TCR:pMHC generalization study — quaidmorris · 2026-07-22
- RSS launches under OMSF to push structural biology data modeling at scale — MoAlQuraishi · 2026-07-22
- enFoldX tops 8 neoantigen scans and an unseen-peptide benchmark — quaidmorris · 2026-07-22
- enFoldX reaches AUC 0.82 on human VDJdb and transfers to mouse at 0.76 — quaidmorris · 2026-07-22