CHANGELOG.md

0.1.0

• Rework of layer API. Geometry inputs are now tensors instead of attrib.
• Tensorflow 2.0 support
• Eager execution
• Build process is now based on tensorflow custom-op docker iamge
• Batch dimension > 1 support

0.0.4

• Added shape_inference::ShapeHandle; to allow Tensorflow to evaluate the shape of the propived operators
• Batch dimension is now handled correctly
• All Operators require the batch_size stored in the first dimension of the Tensor ! Batch dimensions is now mandatory
• Currently all operators are only supporting a batch_size of 1 ! A Batch_size >1 will lead to an Error.
• Preparation to Tensorflow 2.0 and Eager execution are included

0.0.3

• Final license: Apache 2.0

0.0.2

• Critical Bugfix: 3D Layers missing Attributes for gradient registration

0.0.1

• Initial package setup:
• 2D parallel, fan and 3D cone beam projectors and back-projectors computing A and A^T on the fly
• Tensorflow patch for Tensorflow 1.9 up to Tensorflow 1.12
• BUILD file to include custom kernels into the Tensorflow .whl package
• Add license