Align Human Camouflaged Perception: Visual Refocus Reinforcement Fine-Tuning

Ruolin Shen, Xiaozhong Ji, Kai WU, Qingdong He, Jiangning Zhang, Yijun He, Yanghaihua, Xiaobin Hu, xiaoyu sun

Keywords: Human Camouflaged Perception

Abstract:

Current multi-modal models exhibit a notable misalignment with the human visual system when identifying objects that are visually assimilated into the background. Our observations reveal that these multi-modal models cannot distinguish concealed objects, demonstrating an inability to emulate human cognitive processes that effectively utilize foreground-background similarity principles for visual analysis.
To analyze this hidden human-model visual thinking discrepancy, we build a visual system that mimicks human visual camouflaged perception to progressively and iteratively `refocus' visual concealed content. The refocus is a progressive guidance mechanism enabling models to logically localize objects in visual images through stepwise reasoning. The localization process of concealed objects requires hierarchical attention shifting with dynamic adjustment and refinement of prior cognitive knowledge.
In this paper, we propose a visual refocus reinforcement framework via the policy optimization algorithm to encourage multi-modal models to think and refocus more before answering, and achieve excellent reasoning abilities to align human camouflaged perception systems.
Our extensive experiments on camouflaged perception successfully demonstrate the emergence of refocus visual phenomena, characterized by multiple reasoning tokens and dynamic adjustment of the detection box. Besides, experimental results on both camouflaged object classification and detection tasks exhibit significantly superior performance compared to Supervised Fine-Tuning (SFT) baselines.
Furthermore, our visual refocus system surpass the human camouflaged perception in user study where participants are required to identify the target object. We hope that this work will provide valuable insights for advancing future research in multimodal model development. The implementation code and datasets will be made publicly.

Primary Area: reinforcement learning

Submission Number: 10863