Multimodal Clustering Networks for Self-supervised Learning from Unlabeled Videos

TL;DR

This paper introduces a multimodal self-supervised learning framework that creates a shared embedding space for different modalities, enabling cross-modal retrieval and semantic grouping, demonstrated on large-scale video and text datasets.

Contribution

It extends contrastive learning with a multimodal clustering step, improving semantic understanding and retrieval across modalities in a zero-shot setting.

Findings

Achieves state-of-the-art zero-shot retrieval results

Enables cross-modal retrieval across unseen datasets

Effectively captures semantic similarities in multimodal data

Abstract

Multimodal self-supervised learning is getting more and more attention as it allows not only to train large networks without human supervision but also to search and retrieve data across various modalities. In this context, this paper proposes a self-supervised training framework that learns a common multimodal embedding space that, in addition to sharing representations across different modalities, enforces a grouping of semantically similar instances. To this end, we extend the concept of instance-level contrastive learning with a multimodal clustering step in the training pipeline to capture semantic similarities across modalities. The resulting embedding space enables retrieval of samples across all modalities, even from unseen datasets and different domains. To evaluate our approach, we train our model on the HowTo100M dataset and evaluate its zero-shot retrieval capabilities in two challenging domains, namely text-to-video retrieval, and temporal action localization, showing state-of-the-art results on four different datasets.