{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/video-adverb-retrieval-with-compositional","title":"Video-adverb retrieval with compositional adverb-action embeddings","arxiv_id":"2309.15086","date":"2023-09-26","proceeding":null,"authors":["Thomas Hummel","Otniel-Bogdan Mercea","A. Sophia Koepke","Zeynep Akata"],"abstract":"Retrieving adverbs that describe an action in a video poses a crucial step towards fine-grained video understanding. We propose a framework for video-to-adverb retrieval (and vice versa) that aligns video embeddings with their matching compositional adverb-action text embedding in a joint embedding space. The compositional adverb-action text embedding is learned using a residual gating mechanism, along with a novel training objective consisting of triplet losses and a regression target. Our method achieves state-of-the-art performance on five recent benchmarks for video-adverb retrieval. Furthermore, we introduce dataset splits to benchmark video-adverb retrieval for unseen adverb-action compositions on subsets of the MSR-VTT Adverbs and ActivityNet Adverbs datasets. Our proposed framework outperforms all prior works for the generalisation task of retrieving adverbs from videos for unseen adverb-action compositions. Code and dataset splits are available at https://hummelth.github.io/ReGaDa/.","url_abs":"https://arxiv.org/abs/2309.15086v1","url_pdf":"https://arxiv.org/pdf/2309.15086v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"video-adverb-retrieval-with-compositional","repo_url":"https://github.com/ExplainableML/ReGaDa","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":null,"task_name":"Triplet"},{"task_slug":"video-adverb-retrieval","task_name":"Video-Adverb Retrieval"},{"task_slug":"video-adverb-retrieval-unseen-compositions","task_name":"Video-Adverb Retrieval (Unseen Compositions)"}],"methods":[{"method_slug":"regada","method_name":"ReGaDa"}],"datasets_introduced":[],"methods_introduced":[{"slug":"regada","name":"ReGaDa","full_name":"Residual gating mechanism to compose adverb-action representations"}],"results":[{"leaderboard":"/sota/video-adverb-retrieval-on-air","task":"Video-Adverb Retrieval","dataset":"AIR","model":"ReGaDa","rank_in_archive_order":1,"of":4,"metrics":{"Acc-A":"0.874","mAP M":"0.418","mAP W":"0.704"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-on-activitynet-adverbs","task":"Video-Adverb Retrieval","dataset":"ActivityNet Adverbs","model":"ReGaDa","rank_in_archive_order":1,"of":4,"metrics":{"Acc-A":"0.771","mAP M":"0.175","mAP W":"0.239"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-on-howto100m-adverbs","task":"Video-Adverb Retrieval","dataset":"HowTo100M Adverbs","model":"ReGaDa","rank_in_archive_order":1,"of":4,"metrics":{"Acc-A":"0.817","mAP M":"0.528","mAP W":"0.567"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-on-msr-vtt-adverbs","task":"Video-Adverb Retrieval","dataset":"MSR-VTT Adverbs","model":"ReGaDa","rank_in_archive_order":1,"of":4,"metrics":{"Acc-A":"0.786","mAP M":"0.228","mAP W":"0.378"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-on-vatex-adverbs","task":"Video-Adverb Retrieval","dataset":"VATEX Adverbs","model":"ReGaDa","rank_in_archive_order":1,"of":4,"metrics":{"Acc-A":"0.817","mAP M":"0.113","mAP W":"0.29"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on-1","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"ActivityNet Adverbs","model":"ReGaDa","rank_in_archive_order":1,"of":4,"metrics":{"Acc-A":"58.4"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on-1","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"ActivityNet Adverbs","model":"Pseudo Adverbs","rank_in_archive_order":2,"of":4,"metrics":{"Acc-A":"57"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on-1","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"ActivityNet Adverbs","model":"CLIP","rank_in_archive_order":3,"of":4,"metrics":{"Acc-A":"55.1"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on-1","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"ActivityNet Adverbs","model":"Action Changes (reg)","rank_in_archive_order":4,"of":4,"metrics":{"Acc-A":"53.9"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on-2","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"MSR-VTT Adverbs","model":"ReGaDa","rank_in_archive_order":1,"of":5,"metrics":{"Acc-A":"61"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on-2","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"MSR-VTT Adverbs","model":"Action Changes (reg)","rank_in_archive_order":2,"of":5,"metrics":{"Acc-A":"59"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on-2","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"MSR-VTT Adverbs","model":"CLIP","rank_in_archive_order":3,"of":5,"metrics":{"Acc-A":"57"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on-2","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"MSR-VTT Adverbs","model":"Pseudo Adverbs","rank_in_archive_order":4,"of":5,"metrics":{"Acc-A":"56"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on-2","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"MSR-VTT Adverbs","model":"Action Changes (cls)","rank_in_archive_order":5,"of":5,"metrics":{"Acc-A":"53.7"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"VATEX Adverbs","model":"ReGaDa","rank_in_archive_order":1,"of":5,"metrics":{"Acc-A":"61.7"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"VATEX Adverbs","model":"Action Changes (reg)","rank_in_archive_order":2,"of":5,"metrics":{"Acc-A":"54.9"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"VATEX Adverbs","model":"CLIP","rank_in_archive_order":3,"of":5,"metrics":{"Acc-A":"54.5"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"VATEX Adverbs","model":"Action Changes (cls)","rank_in_archive_order":4,"of":5,"metrics":{"Acc-A":"54.3"},"uses_additional_data":false},{"leaderboard":"/sota/video-adverb-retrieval-unseen-compositions-on","task":"Video-Adverb Retrieval (Unseen Compositions)","dataset":"VATEX Adverbs","model":"Pseudo Adverbs","rank_in_archive_order":5,"of":5,"metrics":{"Acc-A":"53.8"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}