MACS: Multi Domain Adaptation Facilitates Accurate Connectomics Segmentation

This repository contains the implementation of MACS, a deep learning framework for connectomics segmentation that introduces the first ever multi-domain adaptation method for connectomics and achieves robust and accurate segmentation performance across diverse connectomics EM datasets.

git clone https://github.com/abrarrahmanabir/MACS.git
cd MACS

All preprocessed datasets used in this study are publicly available and include the train, validation, and test splits to ensure reproducibility. You can access the full dataset at the following link: https://drive.google.com/file/d/1-tuV1zWgcsaz9tEUDRpsrZPoEj9e_glZ/view?usp=sharing

multidomain.py : This file contains the complete implementation of our proposed approach MACS and the corresponding source code.

train_single.py : This script is used for training each individual source model.

test.py : This script contains the code for evaluating the trained models.

The training process is divided into two main stages: training the individual source models and then training the MACS model.

We have provided a bash script to automate the training for all source domains.

To run, execute the following command in your terminal:

bash unet_train.sh

To train the MACS model, use the provided bash script.

bash run_multi.sh

All trained models will be saved in the ./models/ directory.

We have also provided a complete bash script to automate the evaluation process.

To run the evaluation, execute:

bash test_multi.sh

The evaluation results, including all metrics, will be compiled and saved in the multidomain_all_results.csv file.