TT-VidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining Paper โข 2609.33419 โข Published 13 days ago โข 19 โข 3