Presentation

CLIP2StyleGAN: Unsupervised Extraction of StyleGAN Edit Directions
Event Type
Technical Paper
Presentation Types
In Person
Interest Areas
Research & Education
Recordings
This session WILL NOT be recorded.
Registration Categories
Full Conference Supporter
Full Conference
Exhibitor Additional Full Conference
Exhibitor Full Conference
Keywords
Imaging/Video
TimeThursday, 11 August 20229am - 9:05am PDT
LocationEast Building, Room 1-3
DescriptionWe present CLIP2StyleGAN, a framework that effectively links the pretrained latent spaces of StyleGAN and CLIP. The framework can automatically extract semantically labeled edit directions in StyleGAN by finding, naming, and projecting interesting meaningful edit directions from CLIP space, in a fully unsupervised setup, without additional human guidance.