ReMAP: Restoring the Perceptual Cycle with Reasoning-Time Latent Visual Memory
As multimodal large language models (MLLMs) reason for longer, attention to the initial visual input diminishes, weakening visual grounding. Visual memory reintroduces visual evidence during reasoning. We conduct a controlled analysis of visual memory along three axes: curation, organization, and access. We find that l...
Hao-Wen Jiang, Zhan-Yu Guo, Chen-Wei Wu et al.
· 0 citations