{"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/towards-universal-representation-for-unseen","title":"Towards Universal Representation for Unseen Action Recognition","arxiv_id":"1803.08460","date":"2018-03-22","proceeding":"CVPR 2018 6","authors":["Yi Zhu","Yang Long","Yu Guan","Shawn Newsam","Ling Shao"],"abstract":"Unseen Action Recognition (UAR) aims to recognise novel action categories\nwithout training examples. While previous methods focus on inner-dataset\nseen/unseen splits, this paper proposes a pipeline using a large-scale training\nsource to achieve a Universal Representation (UR) that can generalise to a more\nrealistic Cross-Dataset UAR (CD-UAR) scenario. We first address UAR as a\nGeneralised Multiple-Instance Learning (GMIL) problem and discover\n'building-blocks' from the large-scale ActivityNet dataset using distribution\nkernels. Essential visual and semantic components are preserved in a shared\nspace to achieve the UR that can efficiently generalise to new datasets.\nPredicted UR exemplars can be improved by a simple semantic adaptation, and\nthen an unseen action can be directly recognised using UR during the test.\nWithout further training, extensive experiments manifest significant\nimprovements over the UCF101 and HMDB51 benchmarks.","url_abs":"http://arxiv.org/abs/1803.08460v1","url_pdf":"http://arxiv.org/pdf/1803.08460v1.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-recognition-in-videos","task_name":"Action Recognition"},{"task_slug":"multiple-instance-learning","task_name":"Multiple Instance Learning"},{"task_slug":"action-recognition","task_name":"Temporal Action Localization"},{"task_slug":"zero-shot-action-recognition","task_name":"Zero-Shot Action Recognition"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/action-recognition-in-videos-on-activitynet","task":"Action Recognition","dataset":"ActivityNet","model":"CD-UAR","rank_in_archive_order":15,"of":16,"metrics":{"mAP":"53.8"},"uses_additional_data":false},{"leaderboard":"/sota/action-recognition-in-videos-on-hmdb-51","task":"Action Recognition","dataset":"HMDB-51","model":"CD-UAR","rank_in_archive_order":74,"of":77,"metrics":{"Average accuracy of 3 splits":"51.8"},"uses_additional_data":false},{"leaderboard":"/sota/action-recognition-in-videos-on-ucf101","task":"Action Recognition","dataset":"UCF101","model":"CD-UAR","rank_in_archive_order":87,"of":91,"metrics":{"3-fold Accuracy":"42.5"},"uses_additional_data":false},{"leaderboard":"/sota/zero-shot-action-recognition-on-hmdb51","task":"Zero-Shot Action Recognition","dataset":"HMDB51","model":"UR","rank_in_archive_order":21,"of":29,"metrics":{"Top-1 Accuracy":"24.4"},"uses_additional_data":false},{"leaderboard":"/sota/zero-shot-action-recognition-on-ucf101","task":"Zero-Shot Action Recognition","dataset":"UCF101","model":"UR","rank_in_archive_order":26,"of":35,"metrics":{"Top-1 Accuracy":"17.5"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1803.08460","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}