Is Space-Time Attention All You Need for Video Understanding?

TL;DR

TimeSformer introduces a purely attention-based, convolution-free model for video classification that achieves state-of-the-art accuracy, faster training, and better scalability to longer videos by leveraging spatiotemporal self-attention.

Contribution

The paper proposes a novel Transformer-based architecture for video understanding that outperforms existing methods and explores different self-attention schemes, notably divided attention.

Findings

Achieves state-of-the-art accuracy on Kinetics-400 and Kinetics-600 datasets.

Faster training and higher test efficiency compared to 3D CNNs.

Effective on long video clips over one minute long.

Abstract

We present a convolution-free approach to video classification built exclusively on self-attention over space and time. Our method, named "TimeSformer," adapts the standard Transformer architecture to video by enabling spatiotemporal feature learning directly from a sequence of frame-level patches. Our experimental study compares different self-attention schemes and suggests that "divided attention," where temporal attention and spatial attention are separately applied within each block, leads to the best video classification accuracy among the design choices considered. Despite the radically new design, TimeSformer achieves state-of-the-art results on several action recognition benchmarks, including the best reported accuracy on Kinetics-400 and Kinetics-600. Finally, compared to 3D convolutional networks, our model is faster to train, it can achieve dramatically higher test efficiency (at a small drop in accuracy), and it can also be applied to much longer video clips (over one minute long). Code and models are available at: https://github.com/facebookresearch/TimeSformer.