{"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/learning-latent-super-events-to-detect","title":"Learning Latent Super-Events to Detect Multiple Activities in Videos","arxiv_id":"1712.01938","date":"2017-12-05","proceeding":"CVPR 2018 6","authors":["AJ Piergiovanni","Michael S. Ryoo"],"abstract":"In this paper, we introduce the concept of learning latent super-events from\nactivity videos, and present how it benefits activity detection in continuous\nvideos. We define a super-event as a set of multiple events occurring together\nin videos with a particular temporal organization; it is the opposite concept\nof sub-events. Real-world videos contain multiple activities and are rarely\nsegmented (e.g., surveillance videos), and learning latent super-events allows\nthe model to capture how the events are temporally related in videos. We design\ntemporal structure filters that enable the model to focus on particular\nsub-intervals of the videos, and use them together with a soft attention\nmechanism to learn representations of latent super-events. Super-event\nrepresentations are combined with per-frame or per-segment CNNs to provide\nframe-level annotations. Our approach is designed to be fully differentiable,\nenabling end-to-end learning of latent super-event representations jointly with\nthe activity detector using them. Our experiments with multiple public video\ndatasets confirm that the proposed concept of latent super-event learning\nsignificantly benefits activity detection, advancing the state-of-the-arts.","url_abs":"http://arxiv.org/abs/1712.01938v2","url_pdf":"http://arxiv.org/pdf/1712.01938v2.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":"learning-latent-super-events-to-detect","repo_url":"https://github.com/piergiaj/super-events-cvpr18","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"learning-latent-super-events-to-detect","repo_url":"https://github.com/piergiaj/tgm-icml19","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"action-detection","task_name":"Action Detection"},{"task_slug":"activity-detection","task_name":"Activity Detection"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/action-detection-on-charades","task":"Action Detection","dataset":"Charades","model":"Super-events (RGB+Flow)","rank_in_archive_order":14,"of":16,"metrics":{"mAP":"19.41"},"uses_additional_data":true},{"leaderboard":"/sota/action-detection-on-multi-thumos","task":"Action Detection","dataset":"Multi-THUMOS","model":"I3D + our super-event","rank_in_archive_order":6,"of":8,"metrics":{"mAP":"36.4"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1712.01938","atlas_url":"https://app.syntology.ai/?focus=1712.01938","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}