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_214@linklings.com
SUMMARY:Stitch it in Time: GAN-Based Facial Editing of Real Videos
DESCRIPTION:Technical Papers\n\nStitch it in Time: GAN-Based Facial Editin
 g of Real Videos\n\nTzaban, Mokady, Gal, Bermano, Cohen-Or\n\nThe ability 
 of Generative Adversarial Networks to encode rich semantics within their l
 atent space has been widely adopted for facial image editing. However, rep
 licating their success with videos has proven challenging. Applying StyleG
 AN editing over real videos introduces two main challenges: (i) StyleGAN o
 perates over aligned crops. When editing videos, these crops need to be pa
 sted back into the frame, resulting in a spatial inconsistency. (ii) Video
 s introduce a fundamental barrier to overcome - temporal coherency. To add
 ress the first challenge, we propose a novel stitching-tuning procedure. T
 he generator is carefully tuned to overcome the spatial artifacts at crop 
 borders, resulting in a smooth transition even when difficult backgrounds 
 are involved. Turning to temporal coherence, we propose that this challeng
 e is largely artificial. The source video is already temporally coherent, 
 and deviations arise in part due to the careless treatment of individual c
 omponents in the editing pipeline. We leverage the natural alignment of St
 yleGAN and the tendency of neural networks to learn low-frequency function
 s, and demonstrate that they provide a strongly consistent prior. These co
 mponents are combined in an end-to-end framework for semantic editing of f
 acial videos. We compare our pipeline to the current state-of-the-art and 
 demonstrate significant improvements. Our method produces various meaningf
 ul manipulations and maintain greater spatial and temporal consistency, ev
 en in challenging talking head videos which current methods struggle with.
  Our code will be made publicly available.\n\nRegistration Category: FULL 
 ACCESS, EXPERIENCE PLUS ACCESS, EXPERIENCE ACCESS, TRADE EXHIBITOR\n\nLang
 uage: ENGLISH\n\nFormat: IN-PERSON
URL:https://sa2022.siggraph.org/en/full-program/?id=papers_214&sess=sess15
 3
END:VEVENT
END:VCALENDAR