TT-VidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining Paper • 2609.33419 • Published 12 days ago • 19