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 datadataset``gt_path
: The location of your annotation datadataset``output_path
: The path where your results will be saved todataset``checkpoint_path
: DEPRECATEDdataset``delivr_model_path
: The path of the model you want to retraintraining``epochs
: The amount of epochs you want to train your modeltraining``learning_rate
: The learning rate for your trainingtraining``normalization
: Binary,true
performs intensity based normalization on the raw data,false
nottraining``retrain
: Binary,true
retrains the model underdataset``delivr_model_path
.false
nottraining``tta
: Binary,true
performs test time augmentation,false
nottraining``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 emptynetwork``batch_size
: Batch sizenetwork``num_workers
: Number of workers as laid out here