Hand trajectory fusion improves egocentric video query grounding

Arxiv CS.CV (computer vision)

Paper proposes fusing hand motion cues with video appearance for temporal grounding of natural language queries in first-person video. Method outperforms prior approaches on EgoNLQ benchmark.

hand-trajectory

Arxiv CS.CV (computer vision)·Enmin ZhongCarlos R. del-BlancoFernando JaureguizarNarciso Garc\'ia

8 days ago