Product backlog

[JLrumberger]

Current tile_sizes:

• lizard: mixed
• CoNSeP: 224 x 224
• pannuke: mixed ?
• schuerch: 1440×1920

Things we need to built to get the LMDB dataloaders

Dataset specific stuff

• lizard specific transformation (?)
• segpath specific transformation (resize?)
• schuerch specific transformation (collapse labels)
• pannuke specific transformation (?)
• CoNSeP specific transformations (?)

Generall stuff

• Tiling function that works for all
• Resize function (sample, sample_magnification, target_magnification, interpolation, example is here: https://github.com/angelolab/Nimbus-Inference/blob/983cd9baaa6b5dd1699aad5aef26b4e96b4fc5b1/src/nimbus_inference/utils.py#L501
• LMDB prep function or class that writes the tiles (mandatory: he, semantic_mask, tile_names, optional : instance_mask, immunoflourescence, semantic_mask_finegrained)