Learning by Aligning 2D Skeleton Sequences and Multi-Modality Fusion (ECCV 2024)

2D Skeleton Heatmaps and Multi-Modality Fusion for Fine-Grained Human Activity Understanding

LA2DS describes our novel Self-Supervised algorithm that effectively aligns video sequences in time using 2D skeleton heatmaps. Video sequence alignment involves the temporal matching of frames across two videos to ensure accurate synchronization.

Sequence alignment finds applications in various areas, such as auto-annotation, identifying abnormal activities or events in a video, classifying actions or activities performed in a video, and dividing a video into meaningful segments based on the underlying temporal structure, enabling better understanding and analysis of video data.

DOWNLOAD THE PRE-PRINT VERSION FROM ARXIV
GET THE OFFICIAL IMPLEMENTATION OF OUR ECCV 2024 PAPER