Labelling unlabelled videos from scratch with multi-modal self-supervision

TL;DR

This paper introduces a novel multi-modal self-supervised clustering method for pseudo-labeling unlabelled videos by leveraging audio-visual correspondence, enabling semantic clustering without human annotations and providing benchmark results on standard datasets.

Contribution

It presents the first method for unsupervised video labelling using multi-modal self-supervision and introduces benchmark results for this task.

Findings

Clusters have high semantic overlap with ground truth labels

Unsupervised labelling does not emerge naturally from strong feature encoders

Benchmark results on Kinetics, Kinetics-Sound, VGG-Sound, and AVE datasets

Abstract

A large part of the current success of deep learning lies in the effectiveness of data -- more precisely: labelled data. Yet, labelling a dataset with human annotation continues to carry high costs, especially for videos. While in the image domain, recent methods have allowed to generate meaningful (pseudo-) labels for unlabelled datasets without supervision, this development is missing for the video domain where learning feature representations is the current focus. In this work, we a) show that unsupervised labelling of a video dataset does not come for free from strong feature encoders and b) propose a novel clustering method that allows pseudo-labelling of a video dataset without any human annotations, by leveraging the natural correspondence between the audio and visual modalities. An extensive analysis shows that the resulting clusters have high semantic overlap to ground truth human labels. We further introduce the first benchmarking results on unsupervised labelling of common video datasets Kinetics, Kinetics-Sound, VGG-Sound and AVE.