Everything at Once -- Multi-modal Fusion Transformer for Video Retrieval

Abstract

Multi-modal learning from video data has seen increased attention recently as it allows to train semantically meaningful embeddings without human annotation enabling tasks like zero-shot retrieval and classification. In this work, we present a multi-modal, modality agnostic fusion transformer approach that learns to exchange information between multiple modalities, such as video, audio, and text, and integrate them into a joined multi-modal representation to obtain an embedding that aggregates multi-modal temporal information. We propose to train the system with a combinatorial loss on everything at once, single modalities as well as pairs of modalities, explicitly leaving out any add-ons such as position or modality encoding. At test time, the resulting model can process and fuse any number of input modalities. Moreover, the implicit properties of the transformer allow to process inputs of different lengths. To evaluate the proposed approach, we train the model on the large scale HowTo100M dataset and evaluate the resulting embedding space on four challenging benchmark datasets obtaining state-of-the-art results in zero-shot video retrieval and zero-shot video action localization.

Zorro: the masked multimodal transformer

“…Less common are approaches which learn self-supervised models with multiple modalities at once. One recent work in this direction is [46], which learns representations using audio, video and text. However, to avoid the collapse of the self-supervised loss, they feed the modalities two at a time, increasing the amount of necessary forward passes.…”