BiRQA: Bidirectional Robust Quality Assessment for Images

2026-02-23

AI Summary

This work addresses the longstanding limitations of existing full-reference image quality assessment (FR-IQA) methods—namely, slow inference and poor adversarial robustness—by introducing BiRQA, the first approach to simultaneously achieve high accuracy, real-time inference, and strong adversarial robustness in this task. The core of BiRQA lies in a bidirectional multi-scale pyramid architecture that integrates four efficient and complementary feature representations, enhanced by a bottom-up attention mechanism and a top-down cross-gating scheme. Furthermore, it incorporates a novel anchored adversarial training strategy and an uncertainty-aware gating module. Extensive experiments demonstrate that BiRQA matches or surpasses state-of-the-art performance across five public benchmarks while accelerating inference by approximately 3×. Notably, under unseen white-box attacks on KADID-10k, it substantially improves SROCC from 0.30–0.57 to 0.60–0.84.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Security and Privacy: Large-scale security measurements

Abstract

Full-Reference image quality assessment (FR IQA) is important for image compression, restoration and generative modeling, yet current neural metrics remain slow and vulnerable to adversarial perturbations. We present BiRQA, a compact FR IQA metric model that processes four fast complementary features within a bidirectional multiscale pyramid. A bottom-up attention module injects fine-scale cues into coarse levels through an uncertainty-aware gate, while a top-down cross-gating block routes semantic context back to high resolution. To enhance robustness, we introduce Anchored Adversarial Training, a theoretically grounded strategy that uses clean"anchor"samples and a ranking loss to bound pointwise prediction error under attacks. On five public FR IQA benchmarks BiRQA outperforms or matches the previous state of the art (SOTA) while running ~3x faster than previous SOTA models. Under unseen white-box attacks it lifts SROCC from 0.30-0.57 to 0.60-0.84 on KADID-10k, demonstrating substantial robustness gains. To our knowledge, BiRQA is the only FR IQA model combining competitive accuracy with real-time throughput and strong adversarial resilience.

Problem

• Full-Reference Image Quality Assessment
• Adversarial Robustness
• Computational Efficiency
• Image Quality Metrics

Innovation

• Bidirectional Multiscale Pyramid
• Anchored Adversarial Training
• Full-Reference Image Quality Assessment
• Adversarial Robustness
• Cross-Gating Mechanism