CREST: Curvature-Regulated Event-Centric Sampling for Efficient Long-Video Understanding 事件

CREST: Curvature-Regulated Event-Centric Sampling for Efficient Long-Video Understanding arXiv:2605.09223v2 Announce Type: replace Abstract: Selecting informative frames from long videos is a combinatorial problem that existing methods address either through efficient heuristics without explicit modeling of query-conditioned temporal structure, or through multi stage retrieval pipelines with substantial preprocessing cost. We propose \textbf{CREST}, a training-free frame selection method ground