Abstract: Leveraging the effective visual-text alignment and static generalizability from CLIP, recent video learners adopt CLIP initialization with further regularization or recombination for ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results