Multiview Transformers for Video Recognition
This addresses video understanding for AI applications by providing a more efficient and accurate method, though it is incremental as it builds on existing transformer architectures.
The paper tackles video recognition by modeling multiple spatiotemporal resolutions, introducing Multiview Transformers (MTV) with separate encoders and lateral connections, achieving state-of-the-art results on six datasets and improving with large-scale pretraining.
Video understanding requires reasoning at multiple spatiotemporal resolutions -- from short fine-grained motions to events taking place over longer durations. Although transformer architectures have recently advanced the state-of-the-art, they have not explicitly modelled different spatiotemporal resolutions. To this end, we present Multiview Transformers for Video Recognition (MTV). Our model consists of separate encoders to represent different views of the input video with lateral connections to fuse information across views. We present thorough ablation studies of our model and show that MTV consistently performs better than single-view counterparts in terms of accuracy and computational cost across a range of model sizes. Furthermore, we achieve state-of-the-art results on six standard datasets, and improve even further with large-scale pretraining. Code and checkpoints are available at: https://github.com/google-research/scenic/tree/main/scenic/projects/mtv.