Align Before Adapt: Leveraging Entity-to-Region Alignments for Generalizable Video Action Recognition

Open paper

Benchmark	Model	Rank	Results
action-classification-on-kinetics-400	ALT-L/14	–	Acc@1: 88.1
action-classification-on-kinetics-600	ALT-L/14	–	Top-1 Accuracy: 88.6
action-recognition-in-videos-on-something-something-v2	ALT-B/16	–	Top-1 Accuracy: 68.6