Segmentation

This page gives an overview of the segmentation model ImageDataExtractor used to detect particles, as well as how it is used as a standalone segmentation model outside of a data extraction context.

Overview

ImageDataExtractor uses a segmentation method called BPartIS, which is a Bayesian deep neural network trained to segment particles in electron microscopy images. By default, this is run in Bayesian mode, which affords more accurate segmentations at the cost of some speed.

Usage

>>> import cv2
>>> from imagedataextractor.segment import ParticleSegmenter

>>> segmenter = ParticleSegmenter(bayesian=True, tu=0.0125, n_samples=50, device='cpu')

>>> image = cv2.imread('<path/to/image>')  # PIL can also be used

>>> segmentation, uncertainty, _ = segmenter.segment(image)

• The segmenter can be run in discriminative mode by setting bayesian=False.
• tu is the uncertainty threshold beyond which particles are classed as false positives (Bayesian mode only).
• n_samples is the number of Monte Carlo samples used when performance Bayesian inference during prediction (Bayesian mode only).
• The segmenter can be run on a GPU, by setting device='cuda'.