Retinex-Diffusion: On Controlling Illumination Conditions in Diffusion Models via Retinex Theory

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xing, Xiaoyan, Hu, Vincent Tao, Metzen, Jan Hendrik, Groh, Konrad, Karaoglu, Sezer, Gevers, Theo
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929442816786432
author Xing, Xiaoyan
Hu, Vincent Tao
Metzen, Jan Hendrik
Groh, Konrad
Karaoglu, Sezer
Gevers, Theo
author_facet Xing, Xiaoyan
Hu, Vincent Tao
Metzen, Jan Hendrik
Groh, Konrad
Karaoglu, Sezer
Gevers, Theo
contents This paper introduces a novel approach to illumination manipulation in diffusion models, addressing the gap in conditional image generation with a focus on lighting conditions. We conceptualize the diffusion model as a black-box image render and strategically decompose its energy function in alignment with the image formation model. Our method effectively separates and controls illumination-related properties during the generative process. It generates images with realistic illumination effects, including cast shadow, soft shadow, and inter-reflections. Remarkably, it achieves this without the necessity for learning intrinsic decomposition, finding directions in latent space, or undergoing additional training with new datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2407_20785
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Retinex-Diffusion: On Controlling Illumination Conditions in Diffusion Models via Retinex Theory
Xing, Xiaoyan
Hu, Vincent Tao
Metzen, Jan Hendrik
Groh, Konrad
Karaoglu, Sezer
Gevers, Theo
Computer Vision and Pattern Recognition
This paper introduces a novel approach to illumination manipulation in diffusion models, addressing the gap in conditional image generation with a focus on lighting conditions. We conceptualize the diffusion model as a black-box image render and strategically decompose its energy function in alignment with the image formation model. Our method effectively separates and controls illumination-related properties during the generative process. It generates images with realistic illumination effects, including cast shadow, soft shadow, and inter-reflections. Remarkably, it achieves this without the necessity for learning intrinsic decomposition, finding directions in latent space, or undergoing additional training with new datasets.
title Retinex-Diffusion: On Controlling Illumination Conditions in Diffusion Models via Retinex Theory
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2407.20785