Real and synthetic 2D binary masks of kaolin clay particles
Description
The dataset comprises one original scanning electron microscopy (SEM) image of kaolin clay, 4,929 experimentally segmented binary clay particle masks extracted from that image, and 10,000 synthetically generated binary clay particle masks created using a latent space denoising diffusion probabilistic model (DDPM). The original SEM image is provided as a TIFF file (.tif), and both the segmented and generated particle datasets are provided as NumPy arrays (.npy). Each binary particle mask is centered and stored on an 85 × 85 pixel grid. The segmented particle masks represent experimentally observed clay particle geometries extracted through a zero-shot segmentation pipeline based on the Segment Anything Model (SAM), followed by filtering and post-processing. The generated particle masks were synthesized using a latent space DDPM trained on the segmented database. This dataset is part of the paper titled "Synthesizing Clay Particles from Scanning Electron Microscopy Images via Foundation Vision Models and Latent Space Diffusion" by Sara Karimi, Joel Arun, Tyler J. Oathes, Ranuri S. Dissanayaka Mudiyanselage, and Nikolaos N. Vlassis.
Files
Institutions
- Rutgers, The State University of New JerseyNew Jersey, New Brunswick