The Year in AI Papers
The year/Independent research

Paper 2512.17532

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

Jiaqi Tang, Jianmin Chen, Wei Wei, Xiaogang Xu, Runtao Liu, Xiangyu Wu, Qipeng Xie, Jiafei Wu, Lei Zhang, Qifeng Chen

Published
Dec 2025
Research lab
Independent
Citations
8
GitHub
530 stars

01 In brief

Summary

Robust-R1 is a framework designed to improve the robustness of Multimodal Large Language Models (MLLMs) against real-world visual degradations.

Unlike existing methods that rely on implicit training or adaptation of visual encoders, Robust-R1 explicitly models degradations through structured reasoning chains.

The approach consists of three stages: supervised fine-tuning (SFT) to establish degradation-aware reasoning, reinforcement learning with a reward function to align degradation parameters (type and intensity), and dynamic scaling of reasoning chain length based on degradation intensity.

The authors constructed an 11K dataset from A-OKVQA, synthesizing degradations across four stages (acquisition, transmission, environment, postprocessing) and annotating each sample with structured chains.

Evaluations on R-Bench show state-of-the-art performance across all degradation intensities, and on MMMB, MMStar, and RealWorldQA, Robust-R1 demonstrates superior anti-degradation performance with smaller performance drops under adversarial degradations.

Ablation studies confirm the importance of explicit reasoning, the degradation reward, and the length reward.

02 From the paper

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.