Skip to content

Features

This pages lists the progress of CAREamics towards v1.0.0. We list future development, in particular regarding the inclusion of new algorithms, or interoperatibility bridges. Future version will be indexed based on new algorithm additions.

APIs

  • CAREamist API: user-friendly API for training and prediction.
  • Lightning API: using CAREamics modules in PyTorch Lightning for more control and flexibility.

Algorithms

Only algorithms fully integrated into the CAREamics API are checkmarked.

  • Noise2Void, N2V2, structN2V
  • CARE, Noise2Noise
  • UNet semantic segmentation
  • HDN
  • MicroSplit
  • COSDD
  • cryoCARE

To request algorithms to be added, please contribute to the discussion on Github.

File formats

We do not expect to maintain more file formats, but we provide dependency injection mechanisms to allow you to easily consume your own file formats.

  • numpy arrays in memory
  • TIFF format
  • Zarr format (without OME-NGFF metadata)
  • CZI format
  • Custom format via a simple read function
  • Custom format via a more complex ImageStack implementation
  • OME-NGFF (Zarr with OME-NGFF metadata support)
  • Memory-mapped MRC (cryoCARE)

Features

Training

  • Patch exclusion using a mask
  • Background patch filtering
  • Training from list of files without loading them all in memory
  • Training from NGFF format (Zarr only)
  • Splitting validation from training data
  • Skipping validation (Noise2Void only)

Logging

  • WandB and Tensorboard logging (thanks to PyTorch Lightning)
  • Metrics and validation image saving during training

Prediction

  • Tile-by-tile prediction to disk (for Zarr)

Interoperability

  • CLI
  • napari plugin / stand-alone UI
  • nextflow/nf-core modules
  • HPC examples