Pytorch implmentation of a 3D-DDPM and 3DPGGAN
- DDPM model was able to outperform the 3DPGGAN by a few %
- Further optimization of the 3DPGGAN is needed to improve performance
- Dataset obtained as part of MMWHS consists of 20 labelled CT scans for training and 40 unlabelled CT scans for testing
- Images and labels are divided into folders called "/ct_train_images/" and "/ct_train_labels"
- Originally experimentation was not conditioned on labelled images
- 2d axial slices obtained from 3D image at resolution of 128 x 128