This is the official repository of the publication "Virtual reality empowered deep learning analysis of brain activity". This repository contains the source code to train DELIVR from scratch or retrain the existing cFos detection network.

Requirements can be installed from the requirements.txt, Ranger21 has to be installed according to the Ranger21 github repository.

You can run the training pipeline by python __main__.py [CONFIG_LOCATION] where CONFIG_LOCATION points towards the corresponding configuration file.

The configuration file needs to be adapted to your project. This section gives an overview over the configuration options.

• dataset``raw_path : The location of your raw data
• dataset``gt_path : The location of your annotation data
• dataset``output_path : The path where your results will be saved to
• dataset``checkpoint_path : DEPRECATED
• dataset``delivr_model_path : The path of the model you want to retrain
• training``epochs: The amount of epochs you want to train your model
• training``learning_rate: The learning rate for your training
• training``normalization: Binary, true performs intensity based normalization on the raw data, false not
• training``retrain: Binary, true retrains the model under dataset``delivr_model_path. false not
• training``tta: Binary, true performs test time augmentation, false not
• training``test_list: Can be either the path to a dedicated test set or a list of paths pointing to items in your testset. Will be populated automatically if empty
• network``batch_size: Batch size
• network``num_workers: Number of workers as laid out here