LAVA: Language Audio Vision Alignment for Contrastive Video Pre-Training
This addresses the scalability and cost issues in video perception for researchers and practitioners by offering an incremental improvement in self-supervised pre-training.
The paper tackles the problem of generating video representations without expensive hand-annotated data by proposing LAVA, a self-supervised contrastive learning approach that aligns language, audio, and video modalities. It shows competitive performance with state-of-the-art methods on UCF-101 and HMDB-51 action recognition datasets while using less unlabeled data.
Generating representations of video data is of key importance in advancing the field of machine perception. Most current techniques rely on hand-annotated data, which can be difficult to work with, expensive to generate, and hard to scale. In this work, we propose a novel learning approach based on contrastive learning, LAVA, which is capable of learning joint language, audio, and video representations in a self-supervised manner. We pre-train LAVA on the Kinetics 700 dataset using transformer encoders to learn representations for each modality. We then demonstrate that LAVA performs competitively with the current state-of-the-art self-supervised and weakly-supervised pretraining techniques on UCF-101 and HMDB-51 video action recognition while using a fraction of the unlabeled data.