One Diffusion to Generate Them All

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Le, Duong H., Pham, Tuan, Lee, Sangho, Clark, Christopher, Kembhavi, Aniruddha, Mandt, Stephan, Krishna, Ranjay, Lu, Jiasen
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918057224437760
author Le, Duong H.
Pham, Tuan
Lee, Sangho
Clark, Christopher
Kembhavi, Aniruddha
Mandt, Stephan
Krishna, Ranjay
Lu, Jiasen
author_facet Le, Duong H.
Pham, Tuan
Lee, Sangho
Clark, Christopher
Kembhavi, Aniruddha
Mandt, Stephan
Krishna, Ranjay
Lu, Jiasen
contents 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 allows for multi-view generation, camera pose estimation, and instant personalization using sequential image inputs. Our model takes a straightforward yet effective approach by treating all tasks as frame sequences with varying noise scales during training, allowing any frame to act as a conditioning image at inference time. Our unified training framework removes the need for specialized architectures, supports scalable multi-task training, and adapts smoothly to any resolution, enhancing both generalization and scalability. Experimental results demonstrate competitive performance across tasks in both generation and prediction such as text-to-image, multiview generation, ID preservation, depth estimation and camera pose estimation despite relatively small training dataset. Our code and checkpoint are freely available at https://github.com/lehduong/OneDiffusion
format Preprint
id arxiv_https___arxiv_org_abs_2411_16318
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle One Diffusion to Generate Them All
Le, Duong H.
Pham, Tuan
Lee, Sangho
Clark, Christopher
Kembhavi, Aniruddha
Mandt, Stephan
Krishna, Ranjay
Lu, Jiasen
Computer Vision and Pattern Recognition
Artificial Intelligence
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 allows for multi-view generation, camera pose estimation, and instant personalization using sequential image inputs. Our model takes a straightforward yet effective approach by treating all tasks as frame sequences with varying noise scales during training, allowing any frame to act as a conditioning image at inference time. Our unified training framework removes the need for specialized architectures, supports scalable multi-task training, and adapts smoothly to any resolution, enhancing both generalization and scalability. Experimental results demonstrate competitive performance across tasks in both generation and prediction such as text-to-image, multiview generation, ID preservation, depth estimation and camera pose estimation despite relatively small training dataset. Our code and checkpoint are freely available at https://github.com/lehduong/OneDiffusion
title One Diffusion to Generate Them All
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2411.16318