Expectation-Maximization Contrastive Learning for Compact Video-and-Language Representations
This work addresses the challenge of improving video-and-language retrieval for AI applications, offering an incremental enhancement that can be easily integrated into existing methods.
The paper tackles the problem of suboptimal shared latent spaces and modality gaps in video-and-language representation learning by proposing Expectation-Maximization Contrastive Learning (EMCL), which learns compact representations using a set of bases, resulting in significant performance improvements on text-video retrieval benchmarks across all metrics.
Most video-and-language representation learning approaches employ contrastive learning, e.g., CLIP, to project the video and text features into a common latent space according to the semantic similarities of text-video pairs. However, such learned shared latent spaces are not often optimal, and the modality gap between visual and textual representation can not be fully eliminated. In this paper, we propose Expectation-Maximization Contrastive Learning (EMCL) to learn compact video-and-language representations. Specifically, we use the Expectation-Maximization algorithm to find a compact set of bases for the latent space, where the features could be concisely represented as the linear combinations of these bases. Such feature decomposition of video-and-language representations reduces the rank of the latent space, resulting in increased representing power for the semantics. Extensive experiments on three benchmark text-video retrieval datasets prove that our EMCL can learn more discriminative video-and-language representations than previous methods, and significantly outperform previous state-of-the-art methods across all metrics. More encouragingly, the proposed method can be applied to boost the performance of existing approaches either as a jointly training layer or an out-of-the-box inference module with no extra training, making it easy to be incorporated into any existing methods.