aymuos15 / instance_speed Goto Github PK

brats.py -> Original Version from: https://github.com/rachitsaluja/BraTS-2023-Metrics

brats_optimised -> Soumya's version to speed up the code.

soumya.py -> GPU/Torch version of brats_optimised.

The test.py contains the comparison of all the methods. The panoptica (https://github.com/BrainLesion/panoptica) functions are based on two matches. Naive is one to one and the other is many to one.

TODO:

• GPU Version on Connected Components
• Understand why and where the difference in scores occur between panoptica and the brats scores. There is no difference in the case of only one instance.

NOTES:

• Had to make my own fork of panoptica for some minor changes to get the different matchings working and remove redundant prints.
• Currently, CPU is faster than GPU. I am not sure why.

DATA: Our Inhouse Mets-data with marina's nnunet results.