{"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/step-spatio-temporal-progressive-learning-for","title":"STEP: Spatio-Temporal Progressive Learning for Video Action Detection","arxiv_id":"1904.09288","date":"2019-04-19","proceeding":"CVPR 2019 6","authors":["Xitong Yang","Xiaodong Yang","Ming-Yu Liu","Fanyi Xiao","Larry Davis","Jan Kautz"],"abstract":"In this paper, we propose Spatio-TEmporal Progressive (STEP) action\ndetector---a progressive learning framework for spatio-temporal action\ndetection in videos. Starting from a handful of coarse-scale proposal cuboids,\nour approach progressively refines the proposals towards actions over a few\nsteps. In this way, high-quality proposals (i.e., adhere to action movements)\ncan be gradually obtained at later steps by leveraging the regression outputs\nfrom previous steps. At each step, we adaptively extend the proposals in time\nto incorporate more related temporal context. Compared to the prior work that\nperforms action detection in one run, our progressive learning framework is\nable to naturally handle the spatial displacement within action tubes and\ntherefore provides a more effective way for spatio-temporal modeling. We\nextensively evaluate our approach on UCF101 and AVA, and demonstrate superior\ndetection results. Remarkably, we achieve mAP of 75.0% and 18.6% on the two\ndatasets with 3 progressive steps and using respectively only 11 and 34 initial\nproposals.","url_abs":"http://arxiv.org/abs/1904.09288v1","url_pdf":"http://arxiv.org/pdf/1904.09288v1.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":"step-spatio-temporal-progressive-learning-for","repo_url":"https://github.com/NVlabs/STEP","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":"action-recognition-in-videos","task_name":"Action Recognition"},{"task_slug":"video-action-detection","task_name":"Video Action Detection"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/action-detection-on-ucf101-24","task":"Action Detection","dataset":"UCF101-24","model":"STEP","rank_in_archive_order":8,"of":19,"metrics":{"Frame-mAP 0.5":"75","Video-mAP 0.1":"83.1","Video-mAP 0.2":"76.6"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1904.09288","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}