VATT: Transformers for Multimodal Self-Supervised Learning from Raw Video, Audio and Text

TL;DR

VATT introduces a convolution-free Transformer framework for multimodal self-supervised learning from raw video, audio, and text signals, achieving state-of-the-art results across various downstream tasks without supervised pre-training.

Contribution

The paper presents VATT, a novel multimodal Transformer architecture trained end-to-end with contrastive losses, outperforming ConvNet-based models on multiple benchmarks.

Findings

VATT's vision Transformer achieves top-1 accuracy of 82.1% on Kinetics-400.

VATT's audio Transformer sets a new record with 39.4% mAP on AudioSet.

VATT generalizes well to image classification, reaching 78.7% top-1 accuracy on ImageNet.

Abstract

We present a framework for learning multimodal representations from unlabeled data using convolution-free Transformer architectures. Specifically, our Video-Audio-Text Transformer (VATT) takes raw signals as inputs and extracts multimodal representations that are rich enough to benefit a variety of downstream tasks. We train VATT end-to-end from scratch using multimodal contrastive losses and evaluate its performance by the downstream tasks of video action recognition, audio event classification, image classification, and text-to-video retrieval. Furthermore, we study a modality-agnostic, single-backbone Transformer by sharing weights among the three modalities. We show that the convolution-free VATT outperforms state-of-the-art ConvNet-based architectures in the downstream tasks. Especially, VATT's vision Transformer achieves the top-1 accuracy of 82.1% on Kinetics-400, 83.6% on Kinetics-600, 72.7% on Kinetics-700, and 41.1% on Moments in Time, new records while avoiding supervised pre-training. Transferring to image classification leads to 78.7% top-1 accuracy on ImageNet compared to 64.7% by training the same Transformer from scratch, showing the generalizability of our model despite the domain gap between videos and images. VATT's audio Transformer also sets a new record on waveform-based audio event recognition by achieving the mAP of 39.4% on AudioSet without any supervised pre-training. VATT's source code is publicly available.