search

rauldds / Hyperspectral_CT_Recon

Build a prototyping pipeline to test different hyperspectral reconstruction and segmentation approaches leveraging public data and/or simulated data.
2 stars 0 forks source link

Implement patchifying the data and different net depths #15

Closed luisdavid64 closed 1 year ago

luisdavid64 commented 1 year ago

The slices are in Music2d are 100x100x128, and the volumnes in Music2d are even bigger. I think we need to process our data as patches, otherwise we might run into memory issues

rauldds commented 1 year ago

Explore receptive field of network. Tasks here would be to explore which patch size is better and which network depth is better. This gives us an idea whether local or global features are better for the task of material segmentation.

rauldds commented 1 year ago

Maybe even try different channel sizes as well

rauldds commented 1 year ago

THROWING AWAY EMPTY PATCHES - NOT BACKGROUND - BALANCING PROBLEMS - TARGET PATCHES ZOOMING IN COMBINATION geometric sampler and target sampler

luisdavid64 commented 1 year ago

This has been implemented, patchifying + class balancing