LongSpace: Exploring Long-Horizon Spatial Memory from Perception to Recall in Video

Arxiv CS.CV (computer vision)

The paper proposes LongSpace, a framework for long-horizon video understanding that integrates spatial memory across perception, memory, and recall stages. It introduces a spatial memory bank and hierarchical retrieval mechanism.

multimodalllm

Arxiv CS.CV (computer vision)·Shiqiang LangJing LiuHaoyang HePeiwen SunYuanteng ChenTao LiuLan YangLongteng GuoHonggang Zhang

6 days ago