AdaMML: Adaptive Multi-Modal Learning for Efficient Video Recognition

TL;DR

AdaMML introduces an adaptive framework for multi-modal video recognition that dynamically selects the most relevant modalities per segment, significantly reducing computation while improving accuracy.

Contribution

It proposes a novel adaptive multi-modal learning framework with a policy network that dynamically chooses modalities, enhancing efficiency and performance in video recognition.

Findings

Achieves 35%-55% reduction in computation compared to baseline.

Consistently outperforms state-of-the-art methods in accuracy.

Demonstrates effectiveness across four diverse datasets.

Abstract

Multi-modal learning, which focuses on utilizing various modalities to improve the performance of a model, is widely used in video recognition. While traditional multi-modal learning offers excellent recognition results, its computational expense limits its impact for many real-world applications. In this paper, we propose an adaptive multi-modal learning framework, called AdaMML, that selects on-the-fly the optimal modalities for each segment conditioned on the input for efficient video recognition. Specifically, given a video segment, a multi-modal policy network is used to decide what modalities should be used for processing by the recognition model, with the goal of improving both accuracy and efficiency. We efficiently train the policy network jointly with the recognition model using standard back-propagation. Extensive experiments on four challenging diverse datasets demonstrate that our proposed adaptive approach yields 35%-55% reduction in computation when compared to the traditional baseline that simply uses all the modalities irrespective of the input, while also achieving consistent improvements in accuracy over the state-of-the-art methods.