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:20250110T023313Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241206T144500
DTEND;TZID=Asia/Tokyo:20241206T145600
UID:siggraphasia_SIGGRAPH Asia 2024_sess150_papers_938@linklings.com
SUMMARY:360-degree Human Video Generation with 4D Diffusion Transformer
DESCRIPTION:Technical Papers\n\nRuizhi Shao, Youxin Pang, Zerong Zheng, Ji
 ngxiang Sun, and Yebin Liu (Tsinghua University)\n\nWe present a novel app
 roach for generating 360-degree high-quality, spatio-temporally coherent h
 uman videos from a single image. Our framework combines the strengths of d
 iffusion transformers for capturing global correlations across viewpoints 
 and time, and CNNs for accurate condition injection. The core is a hierarc
 hical 4D transformer architecture that factorizes self-attention across vi
 ews, time steps, and spatial dimensions, enabling efficient modeling of th
 e 4D space. Precise conditioning is achieved by injecting human identity, 
 camera parameters, and temporal signals into the respective transformers. 
 To train this model, we collect a multi-dimensional dataset spanning image
 s, videos, multi-view data, and limited 4D footage, along with a tailored 
 multi-dimensional training strategy. Our approach overcomes the limitation
 s of previous methods based on generative adversarial networks or vanilla 
 diffusion models, which struggle with complex motions, viewpoint changes, 
 and generalization. Through extensive experiments, we demonstrate our meth
 od's ability to synthesize 360-degree realistic, coherent human motion vid
 eos, paving the way for advanced multimedia applications in areas such as 
 virtual reality and animation.\n\nRegistration Category: Full Access, Full
  Access Supporter\n\nLanguage Format: English Language\n\nSession Chair: L
 i-Yi Wei (Adobe Research)
URL:https://asia.siggraph.org/2024/program/?id=papers_938&sess=sess150
END:VEVENT
END:VCALENDAR