Synthetic Microscopic Image Generation Toolbox

an mlapp synthetic data generator; simple code to train and test U-Net models; pseudo-labeling adaptive-segmentation pipeline

Overview

Using the Python SyMBac package as the backbone, this package includes two GUIs for synthetic mother machine image generation to train segmentation models and adaptive segmentation to more accurately segment timeseries utilising an initial segmentation model. The package also includes code for U-Net model training and testing.

Installation

• Download the package.
• Clone and install the Python SyMBac package, in the same root folder for convenience:

git clone https://github.com/georgeoshardo/SyMBac.git && git checkout -b symbac origin/symbac
cd symbac && virtualenv env
source env/bin/activate
python setup.py install

Use

• Update config.json to specify the Python virtualenv path.
• app1.mlapp configures synthetic images to be generated. User gets to tune the physical and optical simulation parameters.
• fine_tuning.mlapp wraps up an adaptive segmentation pipeline to perform domain shift on a time series data set continuously and output segmentation masks.