Real and synthetic 3D models of F50 sand grains

The dataset comprises 1,551 real sand grains obtained from synchrotron microcomputed tomography (SMT) scans and 50,000 synthetically generated sand grain 3D models created using a latent space denoising diffusion algorithm. The 3D surface models are provided as Wavefront OBJ files (.obj), and the original and generated point clouds, representing the 3D coordinates of the sand grain surfaces, can be extracted from the vertices of these surface meshes.

The latent space denoising diffusion algorithm generates realistic synthetic sand grain point clouds that emulate the morphological features of real grains. The data shows that synthetic grains exhibit similar shape, size, and surface characteristics to the real grains, making it useful for studying granular assemblies' behavior in engineering applications. This dataset enables detailed analysis and comparison of real and synthetic sand grains, aiming to facilitate the development and validation of new models for granular material analysis.

This dataset is part of the paper titled "Synthesizing Realistic Sand Assemblies with Denoising Diffusion in Latent Space" (2024) by Nikolaos N. Vlassis, WaiChing Sun, Khalid A. Alshibli, and Richard A. Regueiro.