python -m train -d <DATASET_FOLDER>

Or open ProScanNet_train.ipynb in google colab.

Open playground.ipynb in colab

• [KNOWLEDGE] Get some domain knowledge of MRI images and locate areas of interest
• [DATA] Study and implement data augmentation. We want to call a function augment(batch) for each batch to augment (stretch/tilt/etc.) images before parsing it to the model.
• [DATA] Add function that takes images and returns a train and val split so that ratio of positive labels is equal in train and val set
• [MODEL] Add evaluation functions to compute all scores used in challenge (see webpage).
• [MODEL] Add predict.py script that runs evaluations on provided model checkpoint (nice plots(!?))
• [DEV] Add simple yaml based config file for training runs (BJ)
• [DEV] Add wandb experiment tracking and log training progress (BJ)
• [RESEARCH] Look into pre-trained biomedical imaging models (MedNET) (BJ)

• [TRAIN] Use k-fold cross validation