Presentation

<
Blue Noise for Diffusion Models
SessionConsistent Text-to-Image
DescriptionMost existing diffusion models use Gaussian noise for training. We introduce a novel class of deterministic diffusion models using time-varying noise (i.e., from white to blue noise) to incorporate correlation within images during training. Further, our framework allows introducing correlation across images within a single mini-batch to improve gradient flow.
Authors
Xingchang Huang
Max Planck Institute for Informatics
Corentin Salaun
Max Planck Institute for Informatics
Cristina Vasconcelos
Google DeepMind
Christian Theobalt
Max Planck Institute for Informatics
Cengiz Oztireli
Google Research
University of Cambridge
Gurprit Singh
Max Planck Institute for Informatics
Event Type
Technical Paper
TimeMonday, 29 July 20242:50pm - 3:00pm MDT
LocationMile High 1
ACM Digital Library Journal Papers' PDFs
Conference Papers' PDFs
Session Time & Location
Sunday, 28 July 20246:00pm - 8:45pm MDTBluebird Ballroom
Monday, 29 July 20242:00pm - 3:30pm MDTMile High 1
Interest Areas
Research & Education
Recordings
Livestreamed
Recorded
Keywords
AI
Machine Learning
Rendering
Registration Categories
Full Conference
Full Conference Supporter
Virtual Access
Exhibitor Full Conference
Monday
Next PresentationNext PresentationConsistent Text-to-Image - Interactive Discussion