GuardReasoner-VL: Safeguarding VLMs via Reinforced Reasoning

Kun Wang (Nanyang Technological University) · Junfeng Fang (University of Science and Technology of China) · Xinfeng Li (Nanyang Technological University) · Cheng Wang (Huawei Technologies Ltd.) · Mingzhe Du (Nanyang Technological University) · Yue Liu (NUS) · Shengfang Zhai (Peking University) · Yulin Chen ( National University of Singapore) · Tri Cao (National University of Singapore) · Hongcheng Gao (University of the Chinese Academy of Sciences) · Jiaheng Zhang (National University of Singapore) · Bryan Hooi (National University of Singapore)

cold-startdata augmentationdynamic clipping parameterexploration-exploitation f1 scoreguardreasoner-vllength-aware safety rewardonline reinforcement learningreasoning corpusreasoning-based modelrejection samplingsafety-aware data concatenationsupervised fine-tuning token efficiency vlms

To enhance the safety of VLMs, this paper introduces a novel reasoning-based VLM guard model dubbed GuardReasoner-VL. The core idea is to incentivize the guard model to deliberatively reason before making moderation decisions via online RL. First, we construct GuardReasoner-VLTrain, a reasoning corpus with 123K samples and 631K reasoning steps, spanning text, image, and text-image inputs. Then, based on it, we cold-start our model's reasoning ability via SFT. In addition, we further enhance reasoning regarding moderation through online RL. Concretely, to enhance diversity and difficulty of samples, we conduct rejection sampling followed by data augmentation via the proposed safety-aware data concatenation. Besides, we use a dynamic clipping parameter to encourage exploration in early stages and exploitation in later stages. To balance performance and token efficiency, we design a length-aware safety reward that integrates accuracy, format, and token cost. Extensive experiments demonstrate the superiority of our model. Remarkably, it surpasses the runner-up by 19.27% F1 score on average, as shown in Figure 1. We release data, code, and models (3B/7B) of GuardReasoner-VL: https://github.com/yueliu1999/GuardReasoner-VL.