Robust-R1: Degradation-Aware Reasoning for Robust Visual Understanding

TL;DR

Robust-R1 introduces a structured reasoning framework for multimodal models that explicitly models visual degradations, significantly improving robustness against real-world visual impairments through specialized training and adaptive reasoning strategies.

Contribution

The paper presents Robust-R1, a novel degradation-aware reasoning framework that explicitly models visual degradations, with a new dataset and techniques for enhanced robustness in visual understanding.

Findings

Outperforms all baselines on R-Bench degradation benchmark

Maintains superior performance under multi-intensity adversarial degradations

Achieves state-of-the-art robustness in real-world visual scenarios

Abstract

Multimodal Large Language Models struggle to maintain reliable performance under extreme real-world visual degradations, which impede their practical robustness. Existing robust MLLMs predominantly rely on implicit training/adaptation that focuses solely on visual encoder generalization, suffering from limited interpretability and isolated optimization. To overcome these limitations, we propose Robust-R1, a novel framework that explicitly models visual degradations through structured reasoning chains. Our approach integrates: (i) supervised fine-tuning for degradation-aware reasoning foundations, (ii) reward-driven alignment for accurately perceiving degradation parameters, and (iii) dynamic reasoning depth scaling adapted to degradation intensity. To facilitate this approach, we introduce a specialized 11K dataset featuring realistic degradations synthesized across four critical real-world visual processing stages, each annotated with structured chains connecting degradation parameters, perceptual influence, pristine semantic reasoning chain, and conclusion. Comprehensive evaluations demonstrate state-of-the-art robustness: Robust-R1 outperforms all general and robust baselines on the real-world degradation benchmark R-Bench, while maintaining superior anti-degradation performance under multi-intensity adversarial degradations on MMMB, MMStar, and RealWorldQA.