Stardist

- Major update that introduces 3D object detection with star-convex polyhedra.
- Lots of changes under the hood that include several breaking changes.
- New example notebooks.

PyPI release 0.2.1 contained partially wrong files. This release fixes that.
Also add utility function to export results as ImageJ ROIs.

- Fix bug related to non-maximum suppression

- Enable working with multi-channel input images.
- Add improved non-maximum suppression.
- Raise required version of [csbdeep](https://github.com/csbdeep/csbdeep).

- Enable compiling C++ extension with OpenMP support on Windows

The file `dsb2018.zip` contains the train and test images with associated ground truth labels as used in [our paper](https://arxiv.org/abs/1806.03535). They are a subset of the `stage1_train` images from the Kaggle 2018 Data Science Bowl, which are [available in full](https://data.broadinstitute.org/bbbc/BBBC038/) from the [Broad Bioimage Benchmark Collection](https://data.broadinstitute.org/bbbc/).