SyncTweedies: A General Generative Framework Based on Synchronized Diffusions

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Kim, Jaihoon, Koo, Juil, Yeo, Kyeongmin, Sung, Minhyuk
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909374889328640
author Kim, Jaihoon
Koo, Juil
Yeo, Kyeongmin
Sung, Minhyuk
author_facet Kim, Jaihoon
Koo, Juil
Yeo, Kyeongmin
Sung, Minhyuk
contents We introduce a general framework for generating diverse visual content, including ambiguous images, panorama images, mesh textures, and Gaussian splat textures, by synchronizing multiple diffusion processes. We present exhaustive investigation into all possible scenarios for synchronizing multiple diffusion processes through a canonical space and analyze their characteristics across applications. In doing so, we reveal a previously unexplored case: averaging the outputs of Tweedie's formula while conducting denoising in multiple instance spaces. This case also provides the best quality with the widest applicability to downstream tasks. We name this case SyncTweedies. In our experiments generating visual content aforementioned, we demonstrate the superior quality of generation by SyncTweedies compared to other synchronization methods, optimization-based and iterative-update-based methods.
format Preprint
id arxiv_https___arxiv_org_abs_2403_14370
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle SyncTweedies: A General Generative Framework Based on Synchronized Diffusions
Kim, Jaihoon
Koo, Juil
Yeo, Kyeongmin
Sung, Minhyuk
Computer Vision and Pattern Recognition
We introduce a general framework for generating diverse visual content, including ambiguous images, panorama images, mesh textures, and Gaussian splat textures, by synchronizing multiple diffusion processes. We present exhaustive investigation into all possible scenarios for synchronizing multiple diffusion processes through a canonical space and analyze their characteristics across applications. In doing so, we reveal a previously unexplored case: averaging the outputs of Tweedie's formula while conducting denoising in multiple instance spaces. This case also provides the best quality with the widest applicability to downstream tasks. We name this case SyncTweedies. In our experiments generating visual content aforementioned, we demonstrate the superior quality of generation by SyncTweedies compared to other synchronization methods, optimization-based and iterative-update-based methods.
title SyncTweedies: A General Generative Framework Based on Synchronized Diffusions
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2403.14370