Minor changes, but now only supporting Python 3.6 or later (i.e. dropping Python 2.7 and 3.5). - Fix several deprecated uses of other packages (distutils, scipy, numpy, tifffile, keras) - Other minor changes
0.6.3
Maintenance release to support TensorFlow >= 2.6.0, which (again) depends on a standalone Keras package (see [release notes](https://github.com/tensorflow/tensorflow/releases/tag/v2.6.0)).
Since CSBDeep supports TensorFlow 1.x (which always needed a separate Keras package) and TensorFlow 2.x (which never required a separate Keras package until version 2.6.0), we are forced to change the installation instructions:
- *Unchanged* when using TensorFlow 2.x: `pip install csbdeep` - *Updated* when using TensorFlow 1.x: `pip install "csbdeep[tf1]"`
0.6.2
Maintenance release to additionally support Python 3.9 (compatible with TensorFlow 2.5 or later). - Pinning `h5py < 3.0.0` for Python 3.8 or earlier only (due to model serialization issues, see [CSBDeep 0.6.1](https://github.com/CSBDeep/CSBDeep/releases/tag/0.6.1)). - Pinning `h5py >= 3.0.0` for Python 3.9 or later (wheels not available for `h5py < 3.0.0`, model serialization issues fixed in TensorFlow 2.5 or later).
0.6.1
Bugfix release pinning `h5py < 3.0.0` due to model serialization issues with Keras (cf. tensorflow/tensorflow44467).
0.6.0
Major internal changes to support both TensorFlow 1 and 2.
Also add basic support for pre-trained models to `BaseModel` ([moved from StarDist](https://github.com/mpicbg-csbd/stardist/pull/47)).
0.5.2
Require Keras < 2.4, because later versions of Keras depend on TensorFlow 2.x. Otherwise, there are two small bug fixes.