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:20230103T035306Z
LOCATION:Auditorium\, Level 5\, West Wing
DTSTART;TZID=Asia/Seoul:20221206T100000
DTEND;TZID=Asia/Seoul:20221206T120000
UID:siggraphasia_SIGGRAPH Asia 2022_sess153_papers_130@linklings.com
SUMMARY:Disentangled Image Colorization via Global Anchors
DESCRIPTION:Technical Papers\n\nDisentangled Image Colorization via Global
  Anchors\n\nXia, Hu, Wong, Wang\n\nColorization is multimodal by nature an
 d challenges existing frameworks to achieve colorful and structurally cons
 istent results. Even the sophisticated autoregressive model struggles to m
 aintain long-distance color consistency due to the fragility of sequential
  dependence.\nTo overcome this challenge, we propose a novel colorization 
 framework that disentangles color multimodality and structure consistency 
 through global color anchors, so that both aspects could be learned effect
 ively. Our key insight is that several carefully located anchors could rep
 resent the color distribution of an image approximately, and conditioned o
 n the anchor colors, we can predict the image color in a deterministic man
 ner by leveraging the internal correlation. To this end, we construct a co
 lorization model with dual branches, where the color modeler predicts the 
 color distribution for anchor color representation, and the color generato
 r predicts the pixel colors by referring the sampled anchor colors. Import
 antly, the anchors are located under two principles: color independence an
 d global coverage, which is realized with clustering analysis on the deep 
 color features. To simplify the computation, we creatively adopt soft supe
 rpixel segmentation to reduce the image primitives, which still nicely res
 erves the reversibility to pixel-wise representation. Extensive experiment
 s show that our method achieves notable superiority over various mainstrea
 m frameworks in perceptual quality. Thanks to anchor-based color represent
 ation, our model has the flexibility to support diverse and controllable c
 olorization as well.\n\nRegistration Category: FULL ACCESS, EXPERIENCE PLU
 S ACCESS, EXPERIENCE ACCESS, TRADE EXHIBITOR\n\nLanguage: ENGLISH\n\nForma
 t: IN-PERSON
URL:https://sa2022.siggraph.org/en/full-program/?id=papers_130&sess=sess15
 3
END:VEVENT
END:VCALENDAR