BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Seoul
X-LIC-LOCATION:Asia/Seoul
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:KST
DTSTART:18871231T000000
DTSTART:19881009T020000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20230103T035307Z
LOCATION:Auditorium\, Level 5\, West Wing
DTSTART;TZID=Asia/Seoul:20221206T100000
DTEND;TZID=Asia/Seoul:20221206T120000
UID:siggraphasia_SIGGRAPH Asia 2022_sess153_papers_234@linklings.com
SUMMARY:Motion In-betweening via Two-stage Transformers
DESCRIPTION:Technical Papers\n\nMotion In-betweening via Two-stage Transfo
 rmers\n\nQin, Zheng, Zhou\n\nWe present a deep learning-based framework to
  synthesize motion in-betweening in a two-stage manner. Given some context
  frames and a target frame, the system can generate plausible transitions 
 with variable lengths in a non-autoregressive fashion. The framework consi
 sts of two Transformer Encoder-based networks operating in two stages: in 
 the first stage a Context Transformer is designed to generate rough transi
 tions based on the context and in the second stage a Detail Transformer is
  employed to refine motion details. Compared to existing Transformer-based
  methods which either use a complete Transformer Encoder-Decoder architect
 ure or additional 1D convolutions to generate motion transitions, our fram
 ework achieves superior performance with less trainable parameters by only
  leveraging the Transformer Encoder and masked self-attention mechanism. T
 o enhance the generalization of our transformer-based framework, we furthe
 r introduce Keyframe Positional Encoding and Learned Relative Positional E
 ncoding to make our method robust in synthesizing longer transitions excee
 ding the maximum transition length during training. Our framework is also 
 artist-friendly by supporting full and partial pose constraints within the
  transition, giving artists fine control over the synthesized results. We 
 benchmark our framework on the LAFAN1 dataset, and experiments show that o
 ur method outperforms the current state-of-the-art methods at a large marg
 in (an average of 16% for normal-length sequences and 55% for excessive-le
 ngth sequences). Our method trains faster than the RNN-based method and ac
 hieves a four-time speedup during inference. We implement our framework in
 to a production-ready tool inside an animation authoring software and cond
 uct a pilot study to validate the practical value of our method.\n\nRegist
 ration Category: FULL ACCESS, EXPERIENCE PLUS ACCESS, EXPERIENCE ACCESS, T
 RADE EXHIBITOR\n\nLanguage: ENGLISH\n\nFormat: IN-PERSON
URL:https://sa2022.siggraph.org/en/full-program/?id=papers_234&sess=sess15
 3
END:VEVENT
END:VCALENDAR