Keywords: multimodal, fusion, attention, audiovisual, transformers, video
Abstract: Humans perceive the world by concurrently processing and fusing high-dimensional inputs from multiple modalities such as vision and audio. Machine perception models, in stark contrast, are typically modality-specific and optimised for unimodal benchmarks.
A common approach for building multimodal models is to simply combine multiple of these modality-specific architectures using late-stage fusion of final representations or predictions ('late-fusion').
Instead, we introduce a novel transformer based architecture that uses 'attention bottlenecks' for modality fusion at multiple layers. Compared to traditional pairwise self-attention, these bottlenecks force information between different modalities to pass through a small number of '`bottleneck' latent units, requiring the model to collate and condense the most relevant information in each modality and only share what is necessary. We find that such a strategy improves fusion performance, at the same time reducing computational cost. We conduct thorough ablation studies, and achieve state-of-the-art results on multiple audio-visual classification benchmarks including Audioset, Epic-Kitchens and VGGSound. All code and models will be released.
Code Of Conduct: I certify that all co-authors of this work have read and commit to adhering to the NeurIPS Statement on Ethics, Fairness, Inclusivity, and Code of Conduct.
TL;DR: We propose a new multimodal fusion model for video that exchanges cross-modal information between modalities via a small number of 'attention bottleneck' latents, achieving state of the art results for video classification.
Supplementary Material: pdf
Code: https://github.com/google-research/scenic/tree/main/scenic/projects/mbt
11 Replies
Loading