Test Suite

[maschulz]

Integration test

• test brain module with hcp data, single modality, make sure it spits out images of the right shape / data structure
• extra modalities (same here)

Basic function tests (mostly register.py)

• use 3d grid / 2d lattice in center of 3d image and pass through
• see if upsampling the t1 and the lattice still works with the original wapr coefficients, otherwise marc needs to create a new set of upsampled data
• linear registration to MNI
• non-linear registration to MNI
• inversion of warp field/coefficients
• cross registration (back to original participant) - with and without inversion-on-the-fly
• test for one of the hcp participants plus a different resolution. if participants have differently sized t1 (likely) then test on three participants

Datasets

• tests for CombinedDataset / conditional sampling
• MultiModalInvWarpDataset / MultiModalWarpDataset - test if getitem output is as expected (both with single and multi modalities)
• optional: test collate functions

matching

• match_intensity_cuda / numpy - does it run, return image

augmentation.py

• run with different augmentation setups, create a couple of images, compare by hand, use as them as ground truth.

long term: test on different gpu's and python distributions

[maschulz]

@Tomasz-Niewiadomski fyi, the new T2 images are ready and available in main :)