ElasticDiffusion: Training-free Arbitrary Size Image Generation through Global-Local Content Separation

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Haji-Ali, Moayed, Balakrishnan, Guha, Ordonez, Vicente
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917948961062912
author Haji-Ali, Moayed
Balakrishnan, Guha
Ordonez, Vicente
author_facet Haji-Ali, Moayed
Balakrishnan, Guha
Ordonez, Vicente
contents Diffusion models have revolutionized image generation in recent years, yet they are still limited to a few sizes and aspect ratios. We propose ElasticDiffusion, a novel training-free decoding method that enables pretrained text-to-image diffusion models to generate images with various sizes. ElasticDiffusion attempts to decouple the generation trajectory of a pretrained model into local and global signals. The local signal controls low-level pixel information and can be estimated on local patches, while the global signal is used to maintain overall structural consistency and is estimated with a reference image. We test our method on CelebA-HQ (faces) and LAION-COCO (objects/indoor/outdoor scenes). Our experiments and qualitative results show superior image coherence quality across aspect ratios compared to MultiDiffusion and the standard decoding strategy of Stable Diffusion. Project page: https://elasticdiffusion.github.io/
format Preprint
id arxiv_https___arxiv_org_abs_2311_18822
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle ElasticDiffusion: Training-free Arbitrary Size Image Generation through Global-Local Content Separation
Haji-Ali, Moayed
Balakrishnan, Guha
Ordonez, Vicente
Computer Vision and Pattern Recognition
Diffusion models have revolutionized image generation in recent years, yet they are still limited to a few sizes and aspect ratios. We propose ElasticDiffusion, a novel training-free decoding method that enables pretrained text-to-image diffusion models to generate images with various sizes. ElasticDiffusion attempts to decouple the generation trajectory of a pretrained model into local and global signals. The local signal controls low-level pixel information and can be estimated on local patches, while the global signal is used to maintain overall structural consistency and is estimated with a reference image. We test our method on CelebA-HQ (faces) and LAION-COCO (objects/indoor/outdoor scenes). Our experiments and qualitative results show superior image coherence quality across aspect ratios compared to MultiDiffusion and the standard decoding strategy of Stable Diffusion. Project page: https://elasticdiffusion.github.io/
title ElasticDiffusion: Training-free Arbitrary Size Image Generation through Global-Local Content Separation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2311.18822