Event-STVG：面向长视频事件推理的时空视频定位新基准
原标题：Event-STVG: Spatio-Temporal Video Grounding for Long-Form Event Reasoning
AI 摘要
清华大学THU-KEG团队提出Event-STVG新任务，将时空视频定位从描述引导的定位转向长视频事件推理，要求模型根据事件类别、发生次序和语义角色联合定位事件区间及参与者。团队构建了包含8个领域、4016个事件标注的基准，其中足球领域3516个事件。实验显示最强MLLM在10分钟上下文下仅达5.4% m-vIoU，暴露出重复同类事件混淆、空间定位瓶颈和多角色联合定位困难等问题。
正文节选
Event-STVG: Spatio-Temporal Video Grounding for Long-Form Event Reasoning Abstract Spatio-Temporal Video Grounding (STVG) aims to jointly localize a target in space and time given a natural-language query. Despite continued advances, existing STVG benchmarks remain largely centered on localizing targets through instance-specific descriptions, leaving them largely decoupled from long-form reasoning increasingly emphasized in multimodal large language models (MLLMs). In this paper, we introduce Ev