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

Code

• github.com/facebookresearch/TimeSformer
  OfficialIn paperpytorch★ 1,837
• github.com/open-mmlab/mmaction2
  pytorch★ 4,959
• github.com/towhee-io/towhee
  pytorch★ 3,459
• github.com/The-AI-Summer/self-attention-cv
  pytorch★ 1,215
• github.com/lucidrains/TimeSformer-pytorch
  pytorch★ 727
• github.com/mx-mark/videotransformer-pytorch
  pytorch★ 306
• github.com/jerrywn121/TianChi_AIEarth
  pytorch★ 79
• github.com/davide-coccomini/TimeSformer-Video-Classification
  pytorch★ 42
• github.com/yiyixuxu/TimeSformer-rolled-attention
  pytorch★ 40
• github.com/m-bain/video-transformers
  pytorch★ 24

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.

Tasks

Benchmark Results

Reproductions