Recurring the Transformer for Video Action Recognition

Existing video understanding approaches, such as 3D convolutional neural networks and Transformer-Based methods, usually process the videos in a clip-wise manner; hence huge GPU memory is needed and fixed-length video clips are usually required. To alleviate those issues, we introduce a novel Recurrent Vision Transformer (RViT) framework based on spatial-temporal representation learning to achieve the video action recognition task. Specifically, the proposed RViT is equipped with an attention gate to build interaction between current frame input and previous hidden state, thus aggregating the global level interframe features through the hidden state temporally. RViT is executed recurrently to process a video by giving the current frame and previous hidden state. The RViT can capture both spatial and temporal features because of the attention gate and recurrent execution. Besides, the proposed RViT can work on variant-length video clips properly without requiring large GPU memory thanks to the frame by frame processing flow. Our experiment results demonstrate that RViT can achieve state-of-the-art performance on various datasets for the video recognition task. Specifically, RViT can achieve a top-1 accuracy of 81.5% on Kinetics-400, 92.31% on Jester, 67.9% on Something-Something-V2, and an mAP accuracy of 66.1% on Charades.