Zoom-IQA: Image Quality Assessment with Reliable Region-Aware Reasoning

TL;DR

Zoom-IQA tackles no-reference image quality assessment with reliable, region-aware reasoning by combining Grounded-Rationale-IQA data and self-guided exploration. The two-stage pipeline first learns to ground rationale and zoom actions via supervised fine-tuning on GR-IQA, then optimizes a dynamic cropping policy through reinforcement learning with a KL-Coverage regularizer and progressive re-sampling to prevent mode collapse and data bias. Empirical results show improved score regression robustness, enhanced reasoning reliability, and beneficial guidance for downstream image restoration. The approach demonstrates strong zero-shot generalization and offers a transferable framework for interactive perceptual reasoning in vision-language systems.

Abstract

Image Quality Assessment (IQA) is a long-standing problem in computer vision. Previous methods typically focus on predicting numerical scores without explanation or provide low-level descriptions lacking precise scores. Recent reasoning-based vision language models (VLMs) have shown strong potential for IQA, enabling joint generation of quality descriptions and scores. However, we notice that existing VLM-based IQA methods tend to exhibit unreliable reasoning due to their limited capability of integrating visual and textual cues. In this work, we introduce Zoom-IQA, a VLM-based IQA model to explicitly emulate key cognitive behaviors: uncertainty awareness, region reasoning, and iterative refinement. Specifically, we present a two-stage training pipeline: 1) supervised fine-tuning (SFT) on our Grounded-Rationale-IQA (GR-IQA) dataset to teach the model to ground its assessments in key regions; and 2) reinforcement learning (RL) for dynamic policy exploration, primarily stabilized by our KL-Coverage regularizer to prevent reasoning and scoring diversity collapse, and supported by a Progressive Re-sampling Strategy to mitigate annotation bias. Extensive experiments show that Zoom-IQA achieves improved robustness, explainability, and generalization. The application to downstream tasks, such as image restoration, further demonstrates the effectiveness of Zoom-IQA.

Figures (13)

• Figure 1: (Upper) Current IQA methods are non-interactive, leading to inferior assessments. They either spot only partial flaws (e.g., slightly overexposed or slightly blurred) or make factually incorrect claims (clear and well-lit), resulting in erroneous judgments. Our Zoom-IQA uses interactive, region-aware reasoning: it first hypothesizes flaws (green text), then grounds them by cropping (orange text), and finally verifies the degradation (blue text). This hypothesize-and-verify loop provides a complete and accurate assessment. (Lower) Our model's reasoning outputs also benefit downstream tasks, such as text-guided image restoration with SUPIR yu2024scaling. Our prompt enables a far superior restoration compared to those guided by other IQA methods or SUPIR's default VLM, LLaVA-1.5-13b liu2023visual.
• Figure 2: An overview of our two-stage framework. Stage (1), Grounded Quality Rationale Learning (Sec. \ref{['Sec:cold_start']}), first uses SFT to teach the model how to correctly execute the crop action. Stage (2), Self-Guided Exploration (Sec. \ref{['Sec:grpo']}), then uses RL to let the model learn what to crop, allowing it to discover regions that lead to a deeper understanding of image quality.
• Figure 3: The GR-IQA dataset curation pipeline. It uses (1) Visual Reliance Filtering (VRF) to ensure visual grounding via token probabilities, and (2) Hint-Augmented Consistency Filtering (HACF) to perform sentence-level, hint-based checks for unfaithful text.
• Figure 4: Qualitative comparison of Zoom-IQA with competing methods (Q-insight li2025q, VisualQuality-R1 wu2025visualquality). We highlight: correct descriptions, incorrect descriptions, and the uncertainty-aware reasoning unique to our model.
• Figure 5: Qualitative evaluation of reasoning quality on the image restoration task.