08.05.2020 (0.1.0)
- Bumped version to 0.1.0 and released to PyPI
- Approved PR#13 from @moritzknolle with
custom_vnet
(a Conv3D U-Net) implementation - Approved PR#13 from @moritzknolle with attention layers for
custom_unet
andcustom_vnet
- Removed
keras
from requirements insetup.py
- Removed
keras
andtenserflow
fromrequirements.txt
- Fixed image links in
README.md
to make them work on PyPI website
14.02.2020 (0.0.8)
- Bumped version to 0.0.8
- Added support for TF >= 2.0 and solved issue #14
- From this point forward
keras-unet
will importtf.keras
instead of regularKeras
when TF >= 2.0 is present. - Added
color
param tokeras_unet.utils.plot_imgs
function allowing to change the mask color when plotting on top of orginal image. Thanks @muminoff picking that up - Added some docs scafoldings for
keras_unet.utils
27.01.2020 (0.0.7)
- Modified
custom_unet
to not use a bias when using BatchNorm - Added
SpatialDropout2D
tocustom_unet
. Regular Dropout does not perform as well as Spatial Dropout in CNNs. Compare here. - Added test coverage to the new code.
- Added a
__version__
to__init__.py