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:20260817T171534Z
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:Nikos Athanasiou (Max Planck Institute for Intelligent Systems
 ); Alpár Cseke (Max Planck Institute for Intelligent Systems, Meshcapade);
  Markos Diomataris and Michael J. Black (Max Planck Institute for Intellig
 ent 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 m
 otion\nand a textual description of the desired modification, our goal is 
 to generate\nan edited motion as described by the text. The challenges inc
 lude the lack\nof training data and the design of a model that faithfully 
 edits the source\nmotion. In this paper, we address both these challenges.
  We build a methodology\n to semi-automatically collect a dataset of tripl
 ets in the form of (i) a\nsource motion, (ii) a target motion, and (iii) a
 n edit text, and create the new\nMotionFix dataset. Having access to such 
 data allows us to train a conditional\ndiffusion model, TMED, that takes b
 oth the source motion and the edit text\nas input. We further build variou
 s baselines trained only on text-motion\npairs datasets, and show superior
  performance of our model trained on\ntriplets. We introduce new retrieval
 -based metrics for motion editing, and\nestablish a new benchmark on the e
 valuation set of MotionFix. Our results\nare encouraging, paving the way f
 or further research on fine-grained motion\ngeneration. Code and models wi
 ll be made publicly available.\n\nRegistration Category: Full Access, Full
  Access Supporter\n\nLanguage Format: English Language\n\nSession Chair: J
 ungdam Won (Seoul National University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_114&sess=sess117
END:VEVENT
END:VCALENDAR