Self-Supervised Video Representation Learning with Space-Time Cubic Puzzles
This work addresses the high cost of human labeling in video analysis by developing a self-supervised method for learning spatio-temporal representations, offering a domain-specific advancement in video understanding.
The paper tackles the problem of learning video representations without human labels by introducing a self-supervised task called Space-Time Cubic Puzzles, which trains 3D CNNs to arrange permuted spatio-temporal crops, resulting in improved performance on action recognition tasks, outperforming state-of-the-art 2D CNN-based methods on UCF101 and HMDB51 datasets.
Self-supervised tasks such as colorization, inpainting and zigsaw puzzle have been utilized for visual representation learning for still images, when the number of labeled images is limited or absent at all. Recently, this worthwhile stream of study extends to video domain where the cost of human labeling is even more expensive. However, the most of existing methods are still based on 2D CNN architectures that can not directly capture spatio-temporal information for video applications. In this paper, we introduce a new self-supervised task called as \textit{Space-Time Cubic Puzzles} to train 3D CNNs using large scale video dataset. This task requires a network to arrange permuted 3D spatio-temporal crops. By completing \textit{Space-Time Cubic Puzzles}, the network learns both spatial appearance and temporal relation of video frames, which is our final goal. In experiments, we demonstrate that our learned 3D representation is well transferred to action recognition tasks, and outperforms state-of-the-art 2D CNN-based competitors on UCF101 and HMDB51 datasets.