arXiv 2411.16318
One Diffusion to Generate Them All
By Duong H. Le, Tuan Pham, et al.
Published 2024-11-25
Mindmap
Browse the paper's core ideas, clusters, and relationships in a structured outline.
We introduce OneDiffusion, a versatile, large-scale diffusion model that seamlessly supports bidirectional image synthesis and understanding across diverse tasks. It enables conditional generation from inputs such as text, depth, pose, layout, and semantic maps, while also handling tasks like image deblurring, upscaling, and reverse processes such as depth estimation and segmentation. Additionally, OneDiffusion allo…