This repository contains all code used to train models for each of the 3 objectives described in the project report. Note that training will not run out of the box, original training was performed on a SLURM cluster and some paths are hard coded. I also do not include the data used for training here, if you would like the data or if you have any questions about this code please feel free to reach out via email.
Objective 1)
training can be run with the command
python3 train_fgatir_3dpatch.py --preprocess True --num_of_layers 20 --patchSize 70 --fname P3d70 --outf logs/FGtraining_P3d70_MSE_b64_l20_testAGnoise_nB_PS
Objective 2)
python3 train_fgatir_2dpatch_ricloss.py --preprocess True --num_of_layers 20 --patchSize 50 --fname P2d50 --outf logs/FGtraining_P2d50_MSE_b64_l20_testARnoise_Ricloss_nl75
objective 3)
python3 train_dwi.py --preprocess True --num_of_layers 20 --fname p50c5_3 --outf logs/DWItraining_P50dc5_MSE_b64_l20_nl50_6
Dataset classes include the classes DWIdataset and MRIdataset. Loss function classes are testmse.PS() and optRice.riceML()