{"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/hp-gan-probabilistic-3d-human-motion","title":"HP-GAN: Probabilistic 3D human motion prediction via GAN","arxiv_id":"1711.09561","date":"2017-11-27","proceeding":null,"authors":["Emad Barsoum","John Kender","Zicheng Liu"],"abstract":"Predicting and understanding human motion dynamics has many applications,\nsuch as motion synthesis, augmented reality, security, and autonomous vehicles.\nDue to the recent success of generative adversarial networks (GAN), there has\nbeen much interest in probabilistic estimation and synthetic data generation\nusing deep neural network architectures and learning algorithms.\n  We propose a novel sequence-to-sequence model for probabilistic human motion\nprediction, trained with a modified version of improved Wasserstein generative\nadversarial networks (WGAN-GP), in which we use a custom loss function designed\nfor human motion prediction. Our model, which we call HP-GAN, learns a\nprobability density function of future human poses conditioned on previous\nposes. It predicts multiple sequences of possible future human poses, each from\nthe same input sequence but a different vector z drawn from a random\ndistribution. Furthermore, to quantify the quality of the non-deterministic\npredictions, we simultaneously train a motion-quality-assessment model that\nlearns the probability that a given skeleton sequence is a real human motion.\n  We test our algorithm on two of the largest skeleton datasets: NTURGB-D and\nHuman3.6M. We train our model on both single and multiple action types. Its\npredictive power for long-term motion estimation is demonstrated by generating\nmultiple plausible futures of more than 30 frames from just 10 frames of input.\nWe show that most sequences generated from the same input have more than 50\\%\nprobabilities of being judged as a real human sequence. We will release all the\ncode used in this paper to Github.","url_abs":"http://arxiv.org/abs/1711.09561v1","url_pdf":"http://arxiv.org/pdf/1711.09561v1.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":"hp-gan-probabilistic-3d-human-motion","repo_url":"https://github.com/Djmcflush/Funky-Movement","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"hp-gan-probabilistic-3d-human-motion","repo_url":"https://github.com/ThomasDupiereux/hpgan","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"hp-gan-probabilistic-3d-human-motion","repo_url":"https://github.com/ebarsoum/hpgan","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"autonomous-vehicles","task_name":"Autonomous Vehicles"},{"task_slug":"human-pose-forecasting","task_name":"Human Pose Forecasting"},{"task_slug":"human-motion-prediction","task_name":"Human motion prediction"},{"task_slug":"motion-estimation","task_name":"Motion Estimation"},{"task_slug":"motion-synthesis","task_name":"Motion Synthesis"},{"task_slug":"prediction","task_name":"Prediction"},{"task_slug":"synthetic-data-generation","task_name":"Synthetic Data Generation"},{"task_slug":"motion-prediction","task_name":"motion prediction"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/human-pose-forecasting-on-human36m","task":"Human Pose Forecasting","dataset":"Human3.6M","model":"HP-GAN","rank_in_archive_order":28,"of":33,"metrics":{"ADE":"858","APD":"7214","FDE":"867","MMADE":"847","MMFDE":"858"},"uses_additional_data":false},{"leaderboard":"/sota/human-pose-forecasting-on-humaneva-i","task":"Human Pose Forecasting","dataset":"HumanEva-I","model":"HP-GAN","rank_in_archive_order":9,"of":11,"metrics":{"ADE@2000ms":"772","APD@2000ms":"1139","FDE@2000ms":"749"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1711.09561","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}