ChronoReplay：单一答案接口不足以支撑流式空间智能
原标题：Paper page - ChronoReplay: One Answer Interface Is Not Enough for Streaming Spatial Intelligence
AI 摘要
该论文提出 ChronoReplay 训练框架，用于缓解流式空间视觉语言模型（VLM）中的“接口锁定”问题——即模型过度专精于某一种空间答案格式（如类别关系、度量距离或运动轨迹），导致跨任务迁移能力弱。ChronoReplay 将异构空间监督组织为按时间排序的片段，保留各任务原生答案格式并平衡不同监督族的学习。单一检查点在 OVO-S-Bench、OST-Bench 和 UCS-Bench 上同时提升，其中 OST-Bench 提升 13.41 分，并可通过自适应帧选择在推理时以更低视觉成本获得额外增益。
正文节选
ChronoReplay: One Answer Interface Is Not Enough for Streaming Spatial Intelligence Abstract ChronoReplay mitigates interface lock-in in spatial vision-language models by replaying heterogeneous spatial supervision chronologically, improving cross-task transfer and enabling efficient inference-time evidence selection. Training a spatial VLM to answer which side an object lies on does not require it to preserve the metric displacement that determines that answer, while supervision for a distance