Overview

preprocessing

• Modify the save_root and src_root in ./preprocessing/condi-seg/00.baseline_data_clean.py and run as follows. It will convert the nifty images into numpy array and save them in the save_root

python 00.baseline_data_clean.py

train

• Examples of the commandlines for running the 9 experiments in the excel table are listed in ./scripts/condi-seg/. Remember to change the data path in each of the scripts.

• If run on local servers/PCs (Not cs cluster), use sh ./scripts/condi-seg/xxx.sh to train a model. But need to set --use_HPC 0 and --gpu gpu_id. Otherwise the model will only train on the GPU 0.

• If train on cluster. Use the following example command lines if need to submit multiple files. This tool will submit the scripts if its name mactches the regular expressions.

python ./Tools/hpc/qsub.py ./scripts/condi-seg/hpc.01\*

test

• After the training is done, you can copy the models back and do the inference in a local machine to avoid queuing.
• The models will be saved in ./logs/[--project]/[--exp_name]
• Use the following command to do the inference. It will do inference on the best model if the ''[num_epoch]'' is omit.

python test.py ./logs/[--project_name]/[--exp_name] [gpu_id] [num_epoch]