Diffusion Models as Cartoonists: The Curious Case of High Density Regions

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Karczewski, Rafał, Heinonen, Markus, Garg, Vikas
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915309084999680
author Karczewski, Rafał
Heinonen, Markus
Garg, Vikas
author_facet Karczewski, Rafał
Heinonen, Markus
Garg, Vikas
contents We investigate what kind of images lie in the high-density regions of diffusion models. We introduce a theoretical mode-tracking process capable of pinpointing the exact mode of the denoising distribution, and we propose a practical high-density sampler that consistently generates images of higher likelihood than usual samplers. Our empirical findings reveal the existence of significantly higher likelihood samples that typical samplers do not produce, often manifesting as cartoon-like drawings or blurry images depending on the noise level. Curiously, these patterns emerge in datasets devoid of such examples. We also present a novel approach to track sample likelihoods in diffusion SDEs, which remarkably incurs no additional computational cost. Code is available at https://github.com/Aalto-QuML/high-density-diffusion.
format Preprint
id arxiv_https___arxiv_org_abs_2411_01293
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Diffusion Models as Cartoonists: The Curious Case of High Density Regions
Karczewski, Rafał
Heinonen, Markus
Garg, Vikas
Computer Vision and Pattern Recognition
Machine Learning
We investigate what kind of images lie in the high-density regions of diffusion models. We introduce a theoretical mode-tracking process capable of pinpointing the exact mode of the denoising distribution, and we propose a practical high-density sampler that consistently generates images of higher likelihood than usual samplers. Our empirical findings reveal the existence of significantly higher likelihood samples that typical samplers do not produce, often manifesting as cartoon-like drawings or blurry images depending on the noise level. Curiously, these patterns emerge in datasets devoid of such examples. We also present a novel approach to track sample likelihoods in diffusion SDEs, which remarkably incurs no additional computational cost. Code is available at https://github.com/Aalto-QuML/high-density-diffusion.
title Diffusion Models as Cartoonists: The Curious Case of High Density Regions
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2411.01293