登入
洞見 - Egocentric video analysis
Maintaining 3D Spatial Awareness of Active Objects in Egocentric Video
Maintaining the 3D locations of active objects, even when they are out of sight, is critical for correctly locating them over short and long time scales in egocentric video.
Cross-Modal Adaptation of Vision-Language Models for Improved Egocentric Action Recognition
A simple yet effective cross-modal adaptation framework, X-MIC, that injects egocentric video-specific knowledge into the frozen vision-language embedding space, leading to significant improvements in fine-grained cross-dataset recognition of nouns and verbs.
