Fix downscaling in ome.zarr data format

0.3.6

Bump version

Several bug fixes

0.3.5

Bump version

Update inference workflows

0.3.4

Add option to compute 2d filters in watershed

New stats workflow

Add statistics workflow and adapt downscaling workflow to new pybdv syntax

Extend support for inference and rois

0.3.2

Finished debugging blocking with roi

Extend support for label multi-sets; add and improve some functionality

• Extend support for label-multisets to:
  • GraphWorkflow
  • EdgeFeaturesWorfklow
  • FindUniques
• Fix issues in multicut stitching workflow
• Add simple threshold task
• Use pybdv to extend downscaling workflow to bdv.n5 format

New graph storage layout

Change the storage layout for distributed region graphs and distributed edge feature calculation.
Now, n5 varlen chunks are used to store the properties (edges, nodes or features) for each block.

This has the advantage that storage uses less files, because less files are written per block and because the chunks are only stored for non-empty blocks.

However, data from the old storage layout cannot be read any longer with the newer version.

More functionality

• Implement object distance workflow
• Implement multicut based stitching
• Multiscale network inference
• Improve functionality:
  • Various improvements in inference workflow
  • Node label workflow supports label multiset input
  • Add object sizes to region features

Move functionality to elf

• Move general purpose functionality to elf
• Adapt to new nifty.distributed function signatures
• Implement label multiset functionality
• Simplify paintera conversion and downscaling workflow

LMC Paper

Experiments in https://arxiv.org/pdf/1905.10535.pdf were done with this release.