Reproduce “IQA-T1: Tool-based Visual Evidence Reasoning for Image Quality Assessment” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jinjian Wu, Jiaqi Tang, Wei Wei, Yingying Yan, Jianmin Chen, Botong Geng, Lei Zhang, Qifeng Chen
A tool-based visual evidence reasoning framework enhances image quality assessment by integrating explicit, structured visual evidence from specialized perceptual analysis tools into multimodal large language model (MLLM) reasoning. This approach achieves a state-of-the-art average PLCC of 0.795 and SRCC of 0.784 across seven diverse IQA benchmarks, providing interpretable, evidence-grounded quality judgments.
Key insights
- Structured visual evidence generated by external tools effectively mitigates the inherent semantic bias of MLLMs, enhancing their sensitivity to low-level perceptual degradations.
- Dynamic and adaptive invocation of perceptual analysis tools improves IQA performance while maintaining efficiency, outperforming fixed or excessive tool usage.
- A multi-component reward function, including terms for format, scoring accuracy, tool usage, and repetition, is essential for optimizing the model's tool-based reasoning and evidence acquisition strategy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.