Expectation-maximization contrastive learning for compact video-and-language representations
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...
Main Authors: | , , , , , , , |
---|---|
Format: | Conference item |
Language: | English |
Published: |
Curran Associates
2023
|
_version_ | 1797111194807435264 |
---|---|
author | Jin, P Huang, J Liu, F Wu, X Ge, S Song, G Clifton, DA Chen, J |
author_facet | Jin, P Huang, J Liu, F Wu, X Ge, S Song, G Clifton, DA Chen, J |
author_sort | Jin, P |
collection | OXFORD |
description | 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. |
first_indexed | 2024-03-07T08:05:22Z |
format | Conference item |
id | oxford-uuid:9567dbe2-e142-42dd-a050-03f52571b446 |
institution | University of Oxford |
language | English |
last_indexed | 2024-03-07T08:05:22Z |
publishDate | 2023 |
publisher | Curran Associates |
record_format | dspace |
spelling | oxford-uuid:9567dbe2-e142-42dd-a050-03f52571b4462023-10-30T10:10:29ZExpectation-maximization contrastive learning for compact video-and-language representationsConference itemhttp://purl.org/coar/resource_type/c_5794uuid:9567dbe2-e142-42dd-a050-03f52571b446EnglishSymplectic ElementsCurran Associates2023Jin, PHuang, JLiu, FWu, XGe, SSong, GClifton, DAChen, JMost 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. |
spellingShingle | Jin, P Huang, J Liu, F Wu, X Ge, S Song, G Clifton, DA Chen, J Expectation-maximization contrastive learning for compact video-and-language representations |
title | Expectation-maximization contrastive learning for compact video-and-language representations |
title_full | Expectation-maximization contrastive learning for compact video-and-language representations |
title_fullStr | Expectation-maximization contrastive learning for compact video-and-language representations |
title_full_unstemmed | Expectation-maximization contrastive learning for compact video-and-language representations |
title_short | Expectation-maximization contrastive learning for compact video-and-language representations |
title_sort | expectation maximization contrastive learning for compact video and language representations |
work_keys_str_mv | AT jinp expectationmaximizationcontrastivelearningforcompactvideoandlanguagerepresentations AT huangj expectationmaximizationcontrastivelearningforcompactvideoandlanguagerepresentations AT liuf expectationmaximizationcontrastivelearningforcompactvideoandlanguagerepresentations AT wux expectationmaximizationcontrastivelearningforcompactvideoandlanguagerepresentations AT ges expectationmaximizationcontrastivelearningforcompactvideoandlanguagerepresentations AT songg expectationmaximizationcontrastivelearningforcompactvideoandlanguagerepresentations AT cliftonda expectationmaximizationcontrastivelearningforcompactvideoandlanguagerepresentations AT chenj expectationmaximizationcontrastivelearningforcompactvideoandlanguagerepresentations |