This repository contains code and results for segmentation performed on the Cardiac MRI dataset using UNet.
Cardiac MRI contains 30 3D Mono-modal MRI images of the left atrium. 2D patches are generated using medicaltorch. 110 patches with shape (320,320) are generated and transformations are applied on the patches. The list of transformations applied on the training set:
CentreCrop2DElastic TransformationRandom AffineNormalization
After applying these transformations Dataloader is created with batch size of 16. The UNet is then trained untill 30 epochs with intial learning rate of 0.001 and Cosine Annealing Scheduler. Early stopping is also applied with patience of 5 to avoid overfitting.
pytorch == 1.7.0torchvision == 0.8.0medicaltorch
