Tag: Temporal Modeling

TimeSformer cuts video attention cost by 3x vs full spatiotemporal. Compare it to VideoMAE's masked pretraining with real memory benchmarks.

Read more →