Self-Supervised MultiModal Versatile Networks

TL;DR

This paper introduces a versatile self-supervised multi-modal network that integrates visual, audio, and language data from videos, enabling effective representations for various downstream tasks across multiple modalities.

Contribution

The work presents a novel multimodal versatile network capable of combining multiple modalities and applying to diverse tasks, with a new deflation process for static images and videos.

Findings

Achieved state-of-the-art results on UCF101, HMDB51, Kinetics600, AudioSet, and ESC-50.

Demonstrated effective multi-modal representation learning from unlabelled video data.

Network can be applied to video, image, audio, and video-text tasks.

Abstract

Videos are a rich source of multi-modal supervision. In this work, we learn representations using self-supervision by leveraging three modalities naturally present in videos: visual, audio and language streams. To this end, we introduce the notion of a multimodal versatile network -- a network that can ingest multiple modalities and whose representations enable downstream tasks in multiple modalities. In particular, we explore how best to combine the modalities, such that fine-grained representations of the visual and audio modalities can be maintained, whilst also integrating text into a common embedding. Driven by versatility, we also introduce a novel process of deflation, so that the networks can be effortlessly applied to the visual data in the form of video or a static image. We demonstrate how such networks trained on large collections of unlabelled video data can be applied on video, video-text, image and audio tasks. Equipped with these representations, we obtain state-of-the-art performance on multiple challenging benchmarks including UCF101, HMDB51, Kinetics600, AudioSet and ESC-50 when compared to previous self-supervised work. Our models are publicly available.