Improving Multi-Modal Learning with Uni-Modal Teachers

TL;DR

This paper introduces Uni-Modal Teacher, a novel multi-modal learning approach that addresses modality failure by combining fusion objectives with uni-modal distillation, leading to improved individual modality representations and overall task performance.

Contribution

The paper proposes a new multi-modal learning method that mitigates modality failure by integrating uni-modal distillation with fusion objectives, enhancing modality-specific and overall performance.

Findings

Over 3% improvement on VGGSound audio-visual classification.

Significant enhancement in NYU Depth V2 RGB-D image segmentation.

Method generalizes well to various multi-modal fusion approaches.

Abstract

Learning multi-modal representations is an essential step towards real-world robotic applications, and various multi-modal fusion models have been developed for this purpose. However, we observe that existing models, whose objectives are mostly based on joint training, often suffer from learning inferior representations of each modality. We name this problem Modality Failure, and hypothesize that the imbalance of modalities and the implicit bias of common objectives in fusion method prevent encoders of each modality from sufficient feature learning. To this end, we propose a new multi-modal learning method, Uni-Modal Teacher, which combines the fusion objective and uni-modal distillation to tackle the modality failure problem. We show that our method not only drastically improves the representation of each modality, but also improves the overall multi-modal task performance. Our method can be effectively generalized to most multi-modal fusion approaches. We achieve more than 3% improvement on the VGGSound audio-visual classification task, as well as improving performance on the NYU depth V2 RGB-D image segmentation task.