This is a self-contained repository for training regression-based or classification-based, highly-customizable UNET models in Keras. It also lets you train on large datasets with augmentation and without having to load them all directly into memory. All you need is a CSV with file paths to the images.
The 3D Data is DLBS (Dallas Lifespan Brain Study?) T1 images and 6-class tissue segmentation.
You can use ANTsPy to load nifti images (much faster), but it also supports loading from Nibabel.
Scripts to train a model are found in the /code/training/ folder. In particular,
train_segmentation_augment.py shows you how to train a Unet segmentation model with
data augmentation. All it requires is a CSV with file paths.
Additionally, train_AE_augment.py shows you how to train a regression-based Unet
with data augmentation. This script is an autoencoder, but you can easily change
it to predict a different image.
Original image:
Rotated image:
http://theorangeduck.com/page/neural-network-not-working http://www.samcoope.com/posts/machine_learning_research