Simplifying and streamlining large-scale materials image processing with WizardDriven and scalable deep learning

August 05, 2019

Benjamin Provencher (1), Nicolas Piché (1), Mike Marsh (2)
Microscopy and Microanalysis, 25, Supplement 2, August 2019: 402-403. DOI: 10.1017/S1431927619002745


Image methods and imaging throughput--especially in 3D and 4D--continue to evolve and elucidate rich details about materials samples at an ever increasing pace. The full potential for quantitative analysis of those materials, however, is often limited by the speed with which imaging scientists can process the data. For different studies, those bottlenecks can occur both for image analyst time and compute time. To address these rate limitations of image processing, we present an improved Deep Learning engine with a user-guided wizard that simplifies and accelerates user interaction and is parallelized for highthroughput computation on both typical laboratory computer hardware and high-performance compute (HPC) systems. Deep Learning tasks here include, but are not limited to, the fully automatic and parameter-free operations of image denoising and image segmentation.

Author Affiliation

(1) Object Research Systems. Montreal, Canada.
(2) Object Research Systems. Denver, USA.