Fine-grained Image Quality Assessment for Perceptual Image Restoration

2025-08-20

AI Summary

Existing image quality assessment (IQA) metrics struggle to discriminate fine-grained perceptual differences in image restoration (IR) tasks. To address this, we introduce FGRestore—the first fine-grained IQA benchmark specifically designed for IR—featuring paired preference annotations that expose the limitations of conventional IQA metrics in restoration scenarios. Building upon this, we propose FGResQ, a novel deep neural network-based evaluator jointly optimized for coarse-grained score regression and fine-grained pairwise ranking loss. Evaluated across multiple IR tasks, FGResQ consistently outperforms state-of-the-art IQA methods—including LPIPS and DISTS—in both perceptual consistency with human judgment and ranking accuracy. All components—source code, pretrained models, and the FGRestore dataset—are publicly released to advance the field toward more precise, perception-aligned IR quality assessment.

Technology Category

• Computer Vision: Image and Video Retrieval
• Knowledge Representation and Reasoning: Qualitative Reasoning
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Recent years have witnessed remarkable achievements in perceptual image restoration (IR), creating an urgent demand for accurate image quality assessment (IQA), which is essential for both performance comparison and algorithm optimization. Unfortunately, the existing IQA metrics exhibit inherent weakness for IR task, particularly when distinguishing fine-grained quality differences among restored images. To address this dilemma, we contribute the first-of-its-kind fine-grained image quality assessment dataset for image restoration, termed FGRestore, comprising 18,408 restored images across six common IR tasks. Beyond conventional scalar quality scores, FGRestore was also annotated with 30,886 fine-grained pairwise preferences. Based on FGRestore, a comprehensive benchmark was conducted on the existing IQA metrics, which reveal significant inconsistencies between score-based IQA evaluations and the fine-grained restoration quality. Motivated by these findings, we further propose FGResQ, a new IQA model specifically designed for image restoration, which features both coarse-grained score regression and fine-grained quality ranking. Extensive experiments and comparisons demonstrate that FGResQ significantly outperforms state-of-the-art IQA metrics. Codes and model weights have been released in https://pxf0429.github.io/FGResQ/

Problem

• Assessing fine-grained quality differences in restored images
• Addressing limitations of existing image quality assessment metrics
• Developing specialized evaluation for perceptual image restoration tasks

Innovation

• Fine-grained image quality assessment dataset
• Pairwise preference annotations for quality ranking
• Dual-branch model combining score regression and ranking