Skip to content

v0.1.26

Latest
Compare
Choose a tag to compare
@okunator okunator released this 07 May 11:43
· 1 commit to main since this release

0.1.26 — 2025-05-07

Removed

  • Removed datamodules module
  • Removed datasets module

Refactor

  • Refactored the whole model interface to be more user-friendly.

Features

  • Added a new wsi module, including:

    • A SlideReader class to read patches from a WSI slide.
      • Backends: Openslide, CUCIM
      • Adapted the reader class from HistoPrep library. Props to Jopo666
    • get_sub_grids function to get subgrids from a WSI slide. Can be used to filter the patches. Based on connected components.
  • Added a new torch_datasets module, including:

    • WSIDatasetInfer class to run inference directly from WSIs.
      • Adapted the class from HistoPrep library. Props to Jopo666
    • TrainDatasetH5 class to handle training data for the models from a h5 file.
    • TrainDatasetFolder class to handle training data for the models from img and label folders.
  • Added a new inference.WsiSegmenter-class to handle the segmentation of WSIs.

  • Added a new wsi.inst_merger.InstMerger-class to handle the merging of instance masks at image boundaries.

  • Added inst2gdf and sem2gdf functions to utils.vectorize module. These functions convert efficiently instance and semantic masks to GeoDataFrame objects.

  • Added FileHandler.to_mat and FileHandler.to_gson save functions that take in a dictionary of model output masks (output from the Inferer-classes) and save it to a .mat or '.feather', '.geojson', '.parquet' files.

Added Dependencies

  • Added libpysal dependency
  • Added networkx dependency

Removed Dependencies

  • Removed lightning dependency
  • Removed albumentations dependency

Chore

  • Move FolderDatasetInferto torch_datasets module