Self-supervised Video Transformer

In this paper, we propose self-supervised training for video transformers using unlabeled video data. From a given video, we create local and global spatiotemporal views with varying spatial sizes and frame rates. Our self-supervised objective seeks to match the features of these different views representing the same video, to be invariant to spatiotemporal variations in actions. To the best of our knowledge, the proposed approach is the first to alleviate the dependency on negative samples or dedicated memory banks in Self-supervised Video Transformer (SVT). Further, owing to the flexibility of Transformer models, SVT supports slow-fast video processing within a single architecture using dynamically adjusted positional encoding and supports long-term relationship modeling along spatiotemporal dimensions. Our approach performs well on four action recognition benchmarks (Kinetics-400, UCF-101, HMDB-51, and SSv2) and converges faster with small batch sizes. Code: https://git.io/J1juJ

PDF Abstract CVPR 2022 PDF CVPR 2022 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Action Recognition HMDB-51 SVT (finetune) Average accuracy of 3 splits 67.2 # 57
Action Recognition HMDB-51 SVT (linear) Average accuracy of 3 splits 57.8 # 68
Action Classification Kinetics-400 SVT (finetune) Acc@1 78.1 # 121
Action Recognition Something-Something V2 SVT (finetune) Top-1 Accuracy 59.2 # 112
Action Recognition UCF101 SVT (finetune) 3-fold Accuracy 93.7 # 55
Action Recognition UCF101 SVT (linear) 3-fold Accuracy 90.8 # 69

Methods