Attend and Replay: Efficient Action Understanding in Long Videos via Mechanistic Interpretability.

Puyue Hou, Jinjin Zhang, Di Huang

Browse the full ICCV paper archive.