Skip to content

raw2features 0.1.0

Choose a tag to compare

View all tags
@CraigMyles CraigMyles released this 30 Jun 22:52
v0.1.0
6c8e390

First public release of raw2features.

Read whole-slide images in OME-Zarr / OME-NGFF (local or remote) and emit patch- and slide-level foundation-model embeddings as a self-describing .embeddings.zarr store, with the reader, segmenter, patcher, embedder and sink as pluggable seams.

Highlights

  • Cloud-direct reads of NGFF v0.4 and v0.5 (http(s)/s3/gs) - the whole segment -> tile -> embed pipeline runs against a remote store with nothing downloaded.
  • Exact-MPP patch extraction, comparable across slides and scanners.
  • 30+ patch encoders and several slide encoders, with preprocessing transcribed from each model card and weights pinned to a sha256 + Hugging Face revision.
  • Optional GrandQC quality control and stain normalization (Macenko, Reinhard, Vahadane).
  • Multiplex (spatial proteomics) alongside H&E on the same seams, with the marker-aware KRONOS encoder.
  • A specified, self-describing store (JSON Schema + validate-store) and one-way exports to scverse SpatialData, pathology-MIL HDF5, and QuPath GeoJSON.

Install with pip install raw2features (see the README for the optional extras). Full notes in CHANGELOG.md.

Assets 2
Loading