Attention-Based Multimodal Fusion for Video Description

TL;DR

This paper introduces a multimodal attention mechanism for video description that selectively attends to different input modalities, improving the fusion of image, motion, and audio features for more accurate descriptions.

Contribution

The paper proposes a novel multimodal attention mechanism that enhances the fusion of multiple input modalities in video description models.

Findings

Multimodal attention outperforms temporal attention alone.

Achieved competitive results on Youtube2Text dataset.

Significant improvement when combining multimodal and temporal attention.

Abstract

Currently successful methods for video description are based on encoder-decoder sentence generation using recur-rent neural networks (RNNs). Recent work has shown the advantage of integrating temporal and/or spatial attention mechanisms into these models, in which the decoder net-work predicts each word in the description by selectively giving more weight to encoded features from specific time frames (temporal attention) or to features from specific spatial regions (spatial attention). In this paper, we propose to expand the attention model to selectively attend not just to specific times or spatial regions, but to specific modalities of input such as image features, motion features, and audio features. Our new modality-dependent attention mechanism, which we call multimodal attention, provides a natural way to fuse multimodal information for video description. We evaluate our method on the Youtube2Text dataset, achieving results that are competitive with current state of the art. More importantly, we demonstrate that our model incorporating multimodal attention as well as temporal attention significantly outperforms the model that uses temporal attention alone.