GPU Implementation of Random Walks with Prior

This repository contains the GPU versionn of https://github.com/emmanuelle/random_walker/blob/with_data/random_walker.py

helpers: Contains the scripts to run everything. helper_numpy is almost an identical copy paste of the original repo.

visualisation.ipynb: Contains a notebook giving visual unit test.

Performance Results:

GPU Speed up for lapcian @size: (125, 125, 125) 63.231481243642904 % (just run python test.py)

Notes:

• Have to create more unit tests in terms of visualisation

To-do:

• Create a GPU based approach with torch instead of numpy on CPU
• Unit Tests
• Input and Output Testing with mock nifti like files.
• VERY POOR Memory optimisation at the moment.

References :

• The original repo - https://github.com/emmanuelle/random_walker

From the vs runs of aaron

• vsmcrc_102608_0000.nii.gz* -> Raw Volume
• vsmcrc_102608.nii.gz -> Segmentation result from model-1 910-910
• vsmcrc_102608.npz -> Weights for prior from model-1 910-910