This repository provides code for training and evaluating the aNCA model introduced in
Attention Pooling Enhances NCA-based Classification of Microscopy Images.

aNCA extends Neural Cellular Automata (NCA) with an attention pooling mechanism, achieving competitive results on multiple microscopy datasets while maintaining a lightweight architecture.

To set up the required environment, use the provided env_nca.yml file:

conda env create -f env_nca.yml
conda activate env_nca

To train the model, run the following command:

python3 src/train.py --mode train --predict aNCA --output #your_path# --train_set #your_dataset# --fold #your_fold# 

Replace:

• #your_path# with the desired output directory.
• #your_dataset# with your dataset name.
• #your_fold# with a fold number (1-5).

To evaluate the trained model, run:

python3 src/train.py --mode eval --predict aNCA --output #your_path# --train_set #your_dataset# --fold #your_fold# 

Ensure that #your_fold# is one of [1, 2, 3, 4, 5].

To get a summary of the evaluation archive "res.pkl", run:

python3 res.py 

For any questions or issues, feel free to open an issue or reach out! [email protected]