acm-header
Sign In

Communications of the ACM

ACM TechNews

Microscopy Image Segmentation via Point and Shape Regularized Data Synthesis


View as: Print Mobile App Share:
The researchers found their method produced more diverse and realistic images compared to conventional methods.

The segmentation network, trained by point annotations and synthetically generated image-segmentation pairs, automatically segments a real microscopy image into its sought objects (above).

Credit: NYU Tandon School of Engineering

Researchers from New York University's Tandon School of Engineering and Germany's University Hospital Bonn applied a new segmentation network trained by point annotations and synthetically produced image-segmentation pairs to microscopy images.

The framework translates the point annotations into synthetic masks limited by shape information, then uses an advanced generative model to render the masks as realistic microscopic images with consistent object appearance.

The final stage combines the masks and images into a training dataset for a specialized image segmentation model.

The approach yielded more diverse and realistic images than conventional techniques, while supporting linkage between input annotations and generated images when tested on a publicly available dataset.

From NYU Tandon School of Engineering
View Full Article

 

Abstracts Copyright © 2023 SmithBucklin, Washington, D.C., USA


 

No entries found