{"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/actionvlad-learning-spatio-temporal","title":"ActionVLAD: Learning spatio-temporal aggregation for action classification","arxiv_id":"1704.02895","date":"2017-04-10","proceeding":"CVPR 2017 7","authors":["Rohit Girdhar","Deva Ramanan","Abhinav Gupta","Josef Sivic","Bryan Russell"],"abstract":"In this work, we introduce a new video representation for action\nclassification that aggregates local convolutional features across the entire\nspatio-temporal extent of the video. We do so by integrating state-of-the-art\ntwo-stream networks with learnable spatio-temporal feature aggregation. The\nresulting architecture is end-to-end trainable for whole-video classification.\nWe investigate different strategies for pooling across space and time and\ncombining signals from the different streams. We find that: (i) it is important\nto pool jointly across space and time, but (ii) appearance and motion streams\nare best aggregated into their own separate representations. Finally, we show\nthat our representation outperforms the two-stream base architecture by a large\nmargin (13% relative) as well as out-performs other baselines with comparable\nbase architectures on HMDB51, UCF101, and Charades video classification\nbenchmarks.","url_abs":"http://arxiv.org/abs/1704.02895v1","url_pdf":"http://arxiv.org/pdf/1704.02895v1.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":[],"tasks":[{"task_slug":"action-classification","task_name":"Action Classification"},{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"long-video-activity-recognition","task_name":"Long-video Activity Recognition"},{"task_slug":"video-classification","task_name":"Video Classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/long-video-activity-recognition-on-breakfast","task":"Long-video Activity Recognition","dataset":"Breakfast","model":"ActionVlad (I3D-K400-Pretrain-feature)","rank_in_archive_order":8,"of":8,"metrics":{"mAP":"60.20"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1704.02895","atlas_url":"https://app.syntology.ai/?focus=1704.02895","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}