BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250110T023312Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241204T134600
DTEND;TZID=Asia/Tokyo:20241204T135800
UID:siggraphasia_SIGGRAPH Asia 2024_sess117_papers_114@linklings.com
SUMMARY:MotionFix: Text-Driven 3D Human Motion Editing
DESCRIPTION:Technical Papers\n\nNikos Athanasiou (Max Planck Institute for
  Intelligent Systems); Alpár Cseke (Max Planck Institute for Intelligent S
 ystems, Meshcapade); Markos Diomataris and Michael J. Black (Max Planck In
 stitute for Intelligent Systems); and Gül Varol (LIGM,  ́Ecole des Ponts, 
 Univ Gustave Eiffel, CNRS)\n\nThe focus of this paper is 3D motion editing
 . Given a 3D human motion\nand a textual description of the desired modifi
 cation, our goal is to generate\nan edited motion as described by the text
 . The challenges include the lack\nof training data and the design of a mo
 del that faithfully edits the source\nmotion. In this paper, we address bo
 th these challenges. We build a methodology\n to semi-automatically collec
 t a dataset of triplets in the form of (i) a\nsource motion, (ii) a target
  motion, and (iii) an edit text, and create the new\nMotionFix dataset. Ha
 ving access to such data allows us to train a conditional\ndiffusion model
 , TMED, that takes both the source motion and the edit text\nas input. We 
 further build various baselines trained only on text-motion\npairs dataset
 s, and show superior performance of our model trained on\ntriplets. We int
 roduce new retrieval-based metrics for motion editing, and\nestablish a ne
 w benchmark on the evaluation set of MotionFix. Our results\nare encouragi
 ng, paving the way for further research on fine-grained motion\ngeneration
 . Code and models will be made publicly available.\n\nRegistration Categor
 y: Full Access, Full Access Supporter\n\nLanguage Format: English Language
 \n\nSession Chair: Jungdam Won (Seoul National University)
URL:https://asia.siggraph.org/2024/program/?id=papers_114&sess=sess117
END:VEVENT
END:VCALENDAR