Two papers propose improvements to slot-based video object representation

Arxiv CS.CV (computer vision)

TSA (Temporal Slot Activation) and Dual-State Slot Attention both aim to maintain persistent object identity in unsupervised video decomposition. TSA uses a temporal activation mechanism to reallocate slots, while Dual-State decouples appearance and identity to handle occlusions and reappearances.

slot-attentionvideo-object-centric

See 1 more source

Dual-State Slot Attention: Decoupling Appearance and Identity for Video Object-Centric Learning4 days agoSieu Tran, Duc Nguyen, Hao Vo, Khoa Vo, Ngan Le

Arxiv CS.CV (computer vision)·Duc NguyenSieu TranHao VoKhoa VoDuy Minh Ho NguyenNghi D. Q. BuiAnh NguyenLong MaiNgan Le

1 day ago