BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Hong_Kong
X-LIC-LOCATION:Asia/Hong_Kong
BEGIN:STANDARD
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
TZNAME:HKT
DTSTART:19911015T033000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20251218T030656Z
LOCATION:Meeting Room S423+S424\, Level 4
DTSTART;TZID=Asia/Hong_Kong:20251216T105000
DTEND;TZID=Asia/Hong_Kong:20251216T110100
UID:siggraphasia_SIGGRAPH Asia 2025_sess117_papers_1081@linklings.com
SUMMARY:DiffCamera: Arbitrary Refocusing on Images
DESCRIPTION:Yiyang Wang and Xi Chen (The University of Hong Kong), Xiaogan
 g Xu (The Chinese University of Hong Kong), Yu Liu (Tongyi Lab), and Hengs
 huang Zhao (The University of Hong Kong)\n\nThe depth-of-field (DoF) effec
 t, which introduces aesthetically pleasing blur, enhances photographic qua
 lity but is fixed and difficult to modify once the image has been created.
  This becomes problematic when the applied blur is undesirable (e.g., the 
 subject is out of focus).\nTo address this, we propose DiffCamera, a model
  that enables flexible refocusing of a created image conditioned on an arb
 itrary new focus point and a blur level.\nSpecifically, we design a diffus
 ion transformer framework for refocusing learning. However, the training r
 equires pairs of data with different focus planes and bokeh levels in the 
 same scene, which are hard to acquire.\nTo overcome this limitation, we de
 velop a simulation-based pipeline to generate large-scale image pairs with
  varying focus planes and bokeh levels.\nWith the simulated data, we find 
 that training with only a vanilla diffusion objective often leads to incor
 rect DoF behaviors due to the complexity of the task.\nThis requires a str
 onger constraint during training.\nInspired by the photographic principle 
 that photos of different focus planes can be linearly blended into a multi
 -focus image, we propose a stacking constraint during training to enforce 
 precise DoF manipulation.\nThis constraint enhances model training by impo
 sing physically grounded refocusing behavior that the focusing results sho
 uld be faithfully aligned with the scene structure and the camera conditio
 ns so that they can be combined into the correct multi-focus image.\nWe al
 so construct a benchmark to evaluate the effectiveness of our refocusing m
 odel.\nExtensive experiments demonstrate that DiffCamera supports stable r
 efocusing across a wide range of scenes, providing unprecedented control o
 ver DoF adjustments for photography and generative AI applications.\n\nReg
 istration Category: Full Access, Full Access Supporter\n\nSession Chair: Q
 iang Fu (King Abdullah University of Science and Technology (KAUST))\n\n
END:VEVENT
END:VCALENDAR