BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250522T212948Z
LOCATION:Mile High 1
DTSTART;TZID=America/Denver:20240801T090000
DTEND;TZID=America/Denver:20240801T103000
UID:siggraph_SIGGRAPH 2024_sess142@linklings.com
SUMMARY:Controllable Image Generation and Completion
DESCRIPTION:LOOSECONTROL: Lifting ControlNet for Generalized Depth Conditi
 oning\n\nLooseControl introduces a generalized approach for depth-conditio
 ned image generation, overcoming ControlNet's reliance on detailed depth m
 aps. It enables scene creation with boundary and 3D box controls for objec
 t layout. This method simplifies complex environment design, showing promi
 se as a versa...\n\n\nShariq Bhat (King Abdullah University of Science and
  Technology (KAUST)); Niloy Mitra (University College London (UCL), Adobe 
 Research); and Peter Wonka (King Abdullah University of Science and Techno
 logy (KAUST))\n---------------------\nObject-level Scene Deocclusion\n\nIn
  this paper, we present a new self-supervised framework, named PACO, for o
 bject-level scene deocclusion to  deocclude each of the objects of a real-
 world scene. Our approach allows multiple downstream applications, includi
 ng scene-level, single-image 3D reconstruction and object rearrangement in
  i...\n\n\nZhengzhe Liu (The Chinese University of Hong Kong); Qing Liu (A
 dobe Research); Chirui Chang (University of Hong Kong); Jianming Zhang, Da
 niil Pakhomov, Haitian Zheng, and Zhe Lin (Adobe Research); Daniel Cohen-O
 r (Tel Aviv University); and Chi-Wing Fu (The Chinese University of Hong K
 ong)\n---------------------\nControllable Image Generation and Completion 
 - Interactive Discussion\n\nAfter the summary presentations, attendees wil
 l participate in an interactive discussion. Distributed around the room wi
 ll be a series of poster boards for authors to gather around with the audi
 ence. Authors are invited to bring any material related to their paper tha
 t could instigate further conver...\n\n---------------------\nFilter-Guide
 d Diffusion for Controllable Image Generation\n\nFilter-Guided Diffusion (
 FGD) is a controllable, tuning-free, image-to-image translation method for
  diffusion models. It combines fast filtering operations with non-determin
 istic samplers to generate high-quality and diverse images. With its effic
 iency, FGD can be sampled multiple times to outperfor...\n\n\nZeqi Gu (Cor
 nell-Tech Cornell University) and Ethan Yang and Abe Davis (Cornell Univer
 sity)\n---------------------\nRealFill: Reference-driven Generation for Au
 thentic Image Completion\n\nGiven a few reference images that roughly capt
 ure the same scene, and a target image with a missing region, RealFill is 
 able to complete the target image with high-quality image content that is 
 faithful to the true scene.\n\n\nLuming Tang (Cornell University); Natanie
 l Ruiz, Qinghao Chu, Yuanzhen Li, Aleksander Holynski, and David E. Jacobs
  (Google Research); Bharath Hariharan (Cornell University); Yael Pritch an
 d Neal Wadhwa (Google Research); Kfir Aberman (Snap); and Michael Rubinste
 in (Google Research)\n---------------------\nConceptLab: Creative Concept 
 Generation using VLM-Guided Diffusion Prior Constraints\n\nThe surge of pe
 rsonalization techniques has allowed us to imagine how existing concepts w
 ould look in new scenes. However, an intriguing question remains: How can 
 we generate a new, imaginary concept that has never been seen before? We p
 ropose an approach for creative concept generation using Diffus...\n\n\nEl
 ad Richardson, Kfir Goldberg, Yuval Alaluf, and Daniel Cohen-Or (Tel Aviv 
 University) and Yuval Alaluf\n---------------------\nSeparate-and-Enhance:
  Compositional Finetuning for Text-to-image Diffusion Models\n\nThis work 
 targets on improving the compositional capability of text-to-image models.
  Different from previous approaches that requires heavy test-time adaptati
 on per prompt, we propose a compositional finetuning framework with two no
 vel objectives. Through comprehensive evaluations, our model demonst...\n\
 n\nZhipeng Bao (Carnegie Mellon University), Yijun Li and Krishna Kumar Si
 ngh (Adobe Research), Yu-Xiong Wang (University of Illinois Urbana-Champai
 gn), and Martial Hebert (Carnegie Mellon University)\n\nInterest Area: Res
 earch & Education\n\nRegistration Category: Full Conference, Full Conferen
 ce Supporter, Virtual Access, Exhibitor Full Conference, Thursday\n\nSessi
 on Chair: Cecilia Zhang (Adobe, University of California Berkeley)
END:VEVENT
END:VCALENDAR