VidTr: Video Transformer Without Convolutions
arXiv:2104.11746
Abstract
We introduce Video Transformer (VidTr) with separable-attention for video classification. Comparing with commonly used 3D networks, VidTr is able to aggregate spatio-temporal information via stacked attentions and provide better performance with higher efficiency. We first introduce the vanilla video transformer and show that transformer module is able to perform spatio-temporal modeling from raw pixels, but with heavy memory usage. We then present VidTr which reduces the memory cost by 3.3 while keeping the same performance. To further optimize the model, we propose the standard deviation based topK pooling for attention (), which reduces the computation by dropping non-informative features along temporal dimension. VidTr achieves state-of-the-art performance on five commonly used datasets with lower computational requirement, showing both the efficiency and effectiveness of our design. Finally, error analysis and visualization show that VidTr is especially good at predicting actions that require long-term temporal reasoning.
arXiv commentICCV 2021 Accepted
References in corpus (11)
- An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
- UCF101: A Dataset of 101 Human Actions Classes From Videos in The Wild
- Longformer: The Long-Document Transformer
- Is Space-Time Attention All You Need for Video Understanding?
- A Short Note on the Kinetics-700-2020 Human Action Dataset
- Hollywood in Homes: Crowdsourcing Data Collection for Activity Understanding
- CvT: Introducing Convolutions to Vision Transformers
- More Is Less: Learning Efficient Video Representations by Big-Little Network and Depthwise Temporal Aggregation
- Multiscale Vision Transformers
- Video Swin Transformer
- Bridging Text and Video: A Universal Multimodal Transformer for Video-Audio Scene-Aware Dialog