New cGAN reconstructs 3D porous media from 2D images

Researchers have developed a novel conditional Generative Adversarial Network (cGAN) framework capable of reconstructing 3D porous media volumes from 2D images. This method uniquely combines property-conditioned generation with 2D-to-3D reconstruction, bypassing the need for extensive 3D training data while allowing control over petrophysical properties like porosity. The framework utilizes a hybrid architecture with a 3D generator and a 2D discriminator, learning 3D structures from 2D slices and achieving a high $R^2$ of 0.93 for porosity control. AI

IMPACT This research advances generative models for scientific applications, potentially enabling more efficient and cost-effective material property analysis.

RANK_REASON This is a research paper detailing a new methodology for 3D reconstruction using GANs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• dolomite
• dolomite-anhydrite
• generative adversarial network
• U-Net

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →