A General Deep Learning framework for Neuron Instance Segmentation based on Efficient UNet and Morphological Post-processing

Huaqian Wu, Nicolas Souedet, Caroline Jan, Cédric Clouchoux, Thierry Delzescaux

Reference code for the paper A General Deep Learning framework for Neuron Instance Segmentation based on Efficient UNet and Morphological Post-processing. Huaqian Wu, Nicolas Souedet, Caroline Jan, Cédric Clouchoux, Thierry Delzescaux. If you use this code or our datasets, please cite our paper:

@article{wu2022general,
  title={A General Deep Learning framework for Neuron Instance Segmentation based on Efficient UNet and Morphological Post-processing},
  author={Wu, Huaqian and Souedet, Nicolas and Jan, Caroline and Clouchoux, C{\'e}dric and Delzescaux, Thierry},
  journal={arXiv preprint arXiv:2202.08682},
  year={2022}
}

Code

Prerequisite

• Tensorflow
• Keras
• numpy
• tqdm
• OpenCV
• scikit-image

Training

The training set contains the following elements:

• images_all
• masks_all
• folds.csv

Optional: folds.csv attributes each image in the training set a label among [0, 1, 2, 3] for cross validation.

To train a model on a dataset located at ./datasets and save the model weight to ./model, use the following command:

python train_efficient_b5.py -data ./datasets -model ./model

Prediction

To predict images at ./test_data using the model weight at ./model, and save the prediction to ./prediction use the following command:

python predict_efficient_b5.py -i ./test_data -model ./model -o ./prediction

Post-processing

The prediction of the neural network is a probability map, it requires a post-processing step to obtain the final instance segmentation:

python postprocessing.py -i ./prediction/prob.png -o ./prediction/final_segmentation.png -r 10