Towards Fast Adaptation of Pretrained Contrastive Models for Multi-channel Video-Language Retrieval

TL;DR

This paper explores how to quickly adapt contrastive models for multi-channel video-language retrieval, finding that using discrete text tokens with a pretrained contrastive text model offers the best performance and can outperform state-of-the-art methods.

Contribution

It systematically analyzes model design choices for multi-channel video-language retrieval and identifies an effective combination of discrete text tokens with a pretrained contrastive text model.

Findings

Discrete text tokens with contrastive text models outperform other methods.

The proposed approach surpasses state-of-the-art on iVQA and How2QA datasets.

Representing videos as text tokens effectively captures visual information.

Abstract

Multi-channel video-language retrieval require models to understand information from different channels (e.g. video$+$question, video$+$speech) to correctly link a video with a textual response or query. Fortunately, contrastive multimodal models are shown to be highly effective at aligning entities in images/videos and text, e.g., CLIP; text contrastive models are extensively studied recently for their strong ability of producing discriminative sentence embeddings, e.g., SimCSE. However, there is not a clear way to quickly adapt these two lines to multi-channel video-language retrieval with limited data and resources. In this paper, we identify a principled model design space with two axes: how to represent videos and how to fuse video and text information. Based on categorization of recent methods, we investigate the options of representing videos using continuous feature vectors or discrete text tokens; for the fusion method, we explore the use of a multimodal transformer or a pretrained contrastive text model. We extensively evaluate the four combinations on five video-language datasets. We surprisingly find that discrete text tokens coupled with a pretrained contrastive text model yields the best performance, which can even outperform state-of-the-art on the iVQA and How2QA datasets without additional training on millions of video-text data. Further analysis shows that this is because representing videos as text tokens captures the key visual information and text tokens are naturally aligned with text models that are strong retrievers after the contrastive pretraining process. All the empirical analysis establishes a solid foundation for future research on affordable and upgradable multimodal intelligence.