Work

3D clay microstructure synthesis using Denoising Diffusion Probabilistic Models

Year: 2025

Type: article

Abstract: This paper considers the generation of 3D consistent complex microstructures. Our approach transcends the current methods, primarily Generative Adversarial Networks (GANs), by providing a stable training regime that eliminates common artifacts, enhancing the fidelity of generated images, through the utilization of Denoising Diffusion Probabilistic Models (DDPMs). Our DDPMs are found to produce high-quality realizations for two complex clay microstructures, with respect to both diversity and spatial statistics. Furthermore, we show that DDPMs can be implicitely trained to produce porosity-conditioned samples.

Source: Applied Computing and Geosciences

DOI: 10.1016/j.acags.2025.100248

Authors: Ali Aouf, Eric Laloy, Bart Rogiers, Christophe De Vleeschouwer

Institutions: UCLouvain, Belgian Nuclear Research Centre

Language: English

Topic: Generative Adversarial Networks and Image Synthesis

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $2,040

OpenAlex ID: W4410774949

FWCI: 2.156

Cites: 48

Cited by: 6

Related to: 10