DR.Experts: Differential Refinement of Distortion-Aware Experts for Blind Image Quality Assessment

TL;DR

DR.Experts tackles blind image quality assessment by injecting distortion priors into a distortion-aware mixture-of-experts framework. It leverages DA-CLIP to obtain distortion-specific attentions, refines them with a Distortion-Saliency Differential Module to suppress semantic noise, and fuses them with semantic and bridging representations via a Dynamic Distortion Weighting Module. The approach demonstrates improved generalization and data efficiency across five challenging BIQA benchmarks, outperforming state-of-the-art methods on SRCC and PLCC metrics. This work provides a more interpretable, distortion-aware BIQA solution with practical relevance for real-world image quality assessment.

Abstract

Blind Image Quality Assessment, aiming to replicate human perception of visual quality without reference, plays a key role in vision tasks, yet existing models often fail to effectively capture subtle distortion cues, leading to a misalignment with human subjective judgments. We identify that the root cause of this limitation lies in the lack of reliable distortion priors, as methods typically learn shallow relationships between unified image features and quality scores, resulting in their insensitive nature to distortions and thus limiting their performance. To address this, we introduce DR.Experts, a novel prior-driven BIQA framework designed to explicitly incorporate distortion priors, enabling a reliable quality assessment. DR.Experts begins by leveraging a degradation-aware vision-language model to obtain distortion-specific priors, which are further refined and enhanced by the proposed Distortion-Saliency Differential Module through distinguishing them from semantic attentions, thereby ensuring the genuine representations of distortions. The refined priors, along with semantics and bridging representation, are then fused by a proposed mixture-of-experts style module named the Dynamic Distortion Weighting Module. This mechanism weights each distortion-specific feature as per its perceptual impact, ensuring that the final quality prediction aligns with human perception. Extensive experiments conducted on five challenging BIQA benchmarks demonstrate the superiority of DR.Experts over current methods and showcase its excellence in terms of generalization and data efficiency.

Figures (3)

• Figure 1: (a): DR.Experts maintains the advantage even with limited training data. (b): Given a distorted image, the proposed framework DR.Experts first leverages a vision-language model specialized on visual distortions to obtain attention corresponding to various distortions. By differentiating these cues from semantic attention, DR.Experts effectively further purifies distortion-aware representations. These refined features are then adaptively weighted according to their perceptual importance and integrated to yield a precise and perceptually consistent quality assessment.
• Figure 2: Overall architecture of the proposed DR.Experts. We leverage DA-CLIP to obtain priors and use DSDM to refine the attentions. DDWM then, serving as experts, to weigh the importance of distortions and give final predictions.
• Figure 3: Comparison of attention maps from DA-CLIP, the image encoder, and DSDM outputs.