Tencent Text-Video Retrieval: Hierarchical Cross-Modal Interactions with Multi-Level Representations

Text-Video Retrieval plays an important role in multi-modal understanding and has attracted increasing attention in recent years. Most existing methods focus on constructing contrastive pairs between whole videos and complete caption sentences, while overlooking fine-grained cross-modal relationship...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Jiang, Jie, Min, Shaobo, Kong, Weijie, Gong, Dihong, Wang, Hongfa, Li, Zhifeng, Liu, Wei
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Jiang, Jie
Min, Shaobo
Kong, Weijie
Gong, Dihong
Wang, Hongfa
Li, Zhifeng
Liu, Wei
description Text-Video Retrieval plays an important role in multi-modal understanding and has attracted increasing attention in recent years. Most existing methods focus on constructing contrastive pairs between whole videos and complete caption sentences, while overlooking fine-grained cross-modal relationships, e.g., clip-phrase or frame-word. In this paper, we propose a novel method, named Hierarchical Cross-Modal Interaction (HCMI), to explore multi-level cross-modal relationships among video-sentence, clip-phrase, and frame-word for text-video retrieval. Considering intrinsic semantic frame relations, HCMI performs self-attention to explore frame-level correlations and adaptively cluster correlated frames into clip-level and video-level representations. In this way, HCMI constructs multi-level video representations for frame-clip-video granularities to capture fine-grained video content, and multi-level text representations at word-phrase-sentence granularities for the text modality. With multi-level representations for video and text, hierarchical contrastive learning is designed to explore fine-grained cross-modal relationships, i.e., frame-word, clip-phrase, and video-sentence, which enables HCMI to achieve a comprehensive semantic comparison between video and text modalities. Further boosted by adaptive label denoising and marginal sample enhancement, HCMI achieves new state-of-the-art results on various benchmarks, e.g., Rank@1 of 55.0%, 58.2%, 29.7%, 52.1%, and 57.3% on MSR-VTT, MSVD, LSMDC, DiDemo, and ActivityNet, respectively.
doi_str_mv 10.48550/arxiv.2204.03382
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2204_03382</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2204_03382</sourcerecordid><originalsourceid>FETCH-LOGICAL-a672-eb8f010a7e118f013687533b8d35c8a6ed1ed110abd0c93828c73931412716a33</originalsourceid><addsrcrecordid>eNotj8tqwzAQRbXpIqT9gKyqH5ArefxQuiumbQIOhWK6NWNpQgSuHWTVTf--ilsYmAtzOcxhbKNkkuk8lw_oL25O0lRmiQTQ6YrZhgZDQ-ANXYL4cJZG_k7BO5qxf-Q7Rx69OTmDPa_8OE3iMNqY90OIFxPcOEz824UTP3z1wYmaZuoj4expilhcCrfs5oj9RHf_e82al-em2on67XVfPdUCizIV1OmjVBJLUuqaoNBlDtBpC7nRWJBVcWKhs9Js4_falLAFlam0VAUCrNn9H3bRbM_efaL_aa-67aILv3JmUJA</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Tencent Text-Video Retrieval: Hierarchical Cross-Modal Interactions with Multi-Level Representations</title><source>arXiv.org</source><creator>Jiang, Jie ; Min, Shaobo ; Kong, Weijie ; Gong, Dihong ; Wang, Hongfa ; Li, Zhifeng ; Liu, Wei</creator><creatorcontrib>Jiang, Jie ; Min, Shaobo ; Kong, Weijie ; Gong, Dihong ; Wang, Hongfa ; Li, Zhifeng ; Liu, Wei</creatorcontrib><description>Text-Video Retrieval plays an important role in multi-modal understanding and has attracted increasing attention in recent years. Most existing methods focus on constructing contrastive pairs between whole videos and complete caption sentences, while overlooking fine-grained cross-modal relationships, e.g., clip-phrase or frame-word. In this paper, we propose a novel method, named Hierarchical Cross-Modal Interaction (HCMI), to explore multi-level cross-modal relationships among video-sentence, clip-phrase, and frame-word for text-video retrieval. Considering intrinsic semantic frame relations, HCMI performs self-attention to explore frame-level correlations and adaptively cluster correlated frames into clip-level and video-level representations. In this way, HCMI constructs multi-level video representations for frame-clip-video granularities to capture fine-grained video content, and multi-level text representations at word-phrase-sentence granularities for the text modality. With multi-level representations for video and text, hierarchical contrastive learning is designed to explore fine-grained cross-modal relationships, i.e., frame-word, clip-phrase, and video-sentence, which enables HCMI to achieve a comprehensive semantic comparison between video and text modalities. Further boosted by adaptive label denoising and marginal sample enhancement, HCMI achieves new state-of-the-art results on various benchmarks, e.g., Rank@1 of 55.0%, 58.2%, 29.7%, 52.1%, and 57.3% on MSR-VTT, MSVD, LSMDC, DiDemo, and ActivityNet, respectively.</description><identifier>DOI: 10.48550/arxiv.2204.03382</identifier><language>eng</language><subject>Computer Science - Computer Vision and Pattern Recognition</subject><creationdate>2022-04</creationdate><rights>http://creativecommons.org/licenses/by/4.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,777,882</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2204.03382$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2204.03382$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Jiang, Jie</creatorcontrib><creatorcontrib>Min, Shaobo</creatorcontrib><creatorcontrib>Kong, Weijie</creatorcontrib><creatorcontrib>Gong, Dihong</creatorcontrib><creatorcontrib>Wang, Hongfa</creatorcontrib><creatorcontrib>Li, Zhifeng</creatorcontrib><creatorcontrib>Liu, Wei</creatorcontrib><title>Tencent Text-Video Retrieval: Hierarchical Cross-Modal Interactions with Multi-Level Representations</title><description>Text-Video Retrieval plays an important role in multi-modal understanding and has attracted increasing attention in recent years. Most existing methods focus on constructing contrastive pairs between whole videos and complete caption sentences, while overlooking fine-grained cross-modal relationships, e.g., clip-phrase or frame-word. In this paper, we propose a novel method, named Hierarchical Cross-Modal Interaction (HCMI), to explore multi-level cross-modal relationships among video-sentence, clip-phrase, and frame-word for text-video retrieval. Considering intrinsic semantic frame relations, HCMI performs self-attention to explore frame-level correlations and adaptively cluster correlated frames into clip-level and video-level representations. In this way, HCMI constructs multi-level video representations for frame-clip-video granularities to capture fine-grained video content, and multi-level text representations at word-phrase-sentence granularities for the text modality. With multi-level representations for video and text, hierarchical contrastive learning is designed to explore fine-grained cross-modal relationships, i.e., frame-word, clip-phrase, and video-sentence, which enables HCMI to achieve a comprehensive semantic comparison between video and text modalities. Further boosted by adaptive label denoising and marginal sample enhancement, HCMI achieves new state-of-the-art results on various benchmarks, e.g., Rank@1 of 55.0%, 58.2%, 29.7%, 52.1%, and 57.3% on MSR-VTT, MSVD, LSMDC, DiDemo, and ActivityNet, respectively.</description><subject>Computer Science - Computer Vision and Pattern Recognition</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2022</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj8tqwzAQRbXpIqT9gKyqH5ArefxQuiumbQIOhWK6NWNpQgSuHWTVTf--ilsYmAtzOcxhbKNkkuk8lw_oL25O0lRmiQTQ6YrZhgZDQ-ANXYL4cJZG_k7BO5qxf-Q7Rx69OTmDPa_8OE3iMNqY90OIFxPcOEz824UTP3z1wYmaZuoj4expilhcCrfs5oj9RHf_e82al-em2on67XVfPdUCizIV1OmjVBJLUuqaoNBlDtBpC7nRWJBVcWKhs9Js4_falLAFlam0VAUCrNn9H3bRbM_efaL_aa-67aILv3JmUJA</recordid><startdate>20220407</startdate><enddate>20220407</enddate><creator>Jiang, Jie</creator><creator>Min, Shaobo</creator><creator>Kong, Weijie</creator><creator>Gong, Dihong</creator><creator>Wang, Hongfa</creator><creator>Li, Zhifeng</creator><creator>Liu, Wei</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20220407</creationdate><title>Tencent Text-Video Retrieval: Hierarchical Cross-Modal Interactions with Multi-Level Representations</title><author>Jiang, Jie ; Min, Shaobo ; Kong, Weijie ; Gong, Dihong ; Wang, Hongfa ; Li, Zhifeng ; Liu, Wei</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a672-eb8f010a7e118f013687533b8d35c8a6ed1ed110abd0c93828c73931412716a33</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2022</creationdate><topic>Computer Science - Computer Vision and Pattern Recognition</topic><toplevel>online_resources</toplevel><creatorcontrib>Jiang, Jie</creatorcontrib><creatorcontrib>Min, Shaobo</creatorcontrib><creatorcontrib>Kong, Weijie</creatorcontrib><creatorcontrib>Gong, Dihong</creatorcontrib><creatorcontrib>Wang, Hongfa</creatorcontrib><creatorcontrib>Li, Zhifeng</creatorcontrib><creatorcontrib>Liu, Wei</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Jiang, Jie</au><au>Min, Shaobo</au><au>Kong, Weijie</au><au>Gong, Dihong</au><au>Wang, Hongfa</au><au>Li, Zhifeng</au><au>Liu, Wei</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Tencent Text-Video Retrieval: Hierarchical Cross-Modal Interactions with Multi-Level Representations</atitle><date>2022-04-07</date><risdate>2022</risdate><abstract>Text-Video Retrieval plays an important role in multi-modal understanding and has attracted increasing attention in recent years. Most existing methods focus on constructing contrastive pairs between whole videos and complete caption sentences, while overlooking fine-grained cross-modal relationships, e.g., clip-phrase or frame-word. In this paper, we propose a novel method, named Hierarchical Cross-Modal Interaction (HCMI), to explore multi-level cross-modal relationships among video-sentence, clip-phrase, and frame-word for text-video retrieval. Considering intrinsic semantic frame relations, HCMI performs self-attention to explore frame-level correlations and adaptively cluster correlated frames into clip-level and video-level representations. In this way, HCMI constructs multi-level video representations for frame-clip-video granularities to capture fine-grained video content, and multi-level text representations at word-phrase-sentence granularities for the text modality. With multi-level representations for video and text, hierarchical contrastive learning is designed to explore fine-grained cross-modal relationships, i.e., frame-word, clip-phrase, and video-sentence, which enables HCMI to achieve a comprehensive semantic comparison between video and text modalities. Further boosted by adaptive label denoising and marginal sample enhancement, HCMI achieves new state-of-the-art results on various benchmarks, e.g., Rank@1 of 55.0%, 58.2%, 29.7%, 52.1%, and 57.3% on MSR-VTT, MSVD, LSMDC, DiDemo, and ActivityNet, respectively.</abstract><doi>10.48550/arxiv.2204.03382</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2204.03382
ispartof
issn
language eng
recordid cdi_arxiv_primary_2204_03382
source arXiv.org
subjects Computer Science - Computer Vision and Pattern Recognition
title Tencent Text-Video Retrieval: Hierarchical Cross-Modal Interactions with Multi-Level Representations
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-18T20%3A17%3A20IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Tencent%20Text-Video%20Retrieval:%20Hierarchical%20Cross-Modal%20Interactions%20with%20Multi-Level%20Representations&rft.au=Jiang,%20Jie&rft.date=2022-04-07&rft_id=info:doi/10.48550/arxiv.2204.03382&rft_dat=%3Carxiv_GOX%3E2204_03382%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true