3D Convolutional sliding window approach is used to detect growth plate plane in 3D micro CT images in mice.
Needed libraries are listed in environment.yml and can be installed with conda
conda env create -f environment.yml
source setup.sh
can be used to setup the python paths.
Training can be run using the following commant
python scripts/train_3dwindow_CV.py --config configs/config_train.json
Paths in the config should be adapted for the input data paths. The script runs 5-fold CV where folds are defined in the input train_with_folds.csv file.
Initial ResNet34 weights can be downloaded from MedicalNet
After the models are trained the inference can be run with
python scripts/run_inference.py --config configs/config_inference.json
that will create csv file with mean prediction, and json file with prediction for each checkpoint in the trained weights folder.