Audiovisual Masked Autoencoders
This work addresses the challenge of improving representation learning for audiovisual tasks, offering a method that enhances performance across multiple datasets and enables transferability to new domains.
The paper tackles the problem of self-supervised representation learning by leveraging audiovisual information in video, achieving significant improvements on downstream classification tasks and surpassing state-of-the-art results on VGGSound and AudioSet.
Can we leverage the audiovisual information already present in video to improve self-supervised representation learning? To answer this question, we study various pretraining architectures and objectives within the masked autoencoding framework, motivated by the success of similar methods in natural language and image understanding. We show that we can achieve significant improvements on audiovisual downstream classification tasks, surpassing the state-of-the-art on VGGSound and AudioSet. Furthermore, we can leverage our audiovisual pretraining scheme for multiple unimodal downstream tasks using a single audiovisual pretrained model. We additionally demonstrate the transferability of our representations, achieving state-of-the-art audiovisual results on Epic Kitchens without pretraining specifically for this dataset.