· Contributors · Organizations · Search
Palette: Image-to-image Diffusion Models
Research & Education
Full Conference Supporter
Exhibitor Additional Full Conference
Exhibitor Full Conference
This session WILL NOT be recorded.
LocationEast Building, Ballroom A/B
DescriptionWe’ve developed a simple and general framework for image-to-image translation based on iterative refinement, called Palette. We’ve trained and evaluated Palette on four challenging computer vision tasks, namely colorization, inpainting, uncropping, and JPEG restoration. Palette generates photorealistic results and outperforms strong task-specific GANs, without any task-specific customization or hyper-parameter tuning.