Initial Release

This release has the following features:

• General Fake Dataset
• General Losses:
  • Distance Loss
  • Focal Loss (binary/categorical and with/without logits)
  • Soft Dice Loss
• Metrics:
  • Dice Coefficient
• Models:
  • General Nd Unet/Linknet with varying depth and adjustable parameters
• Optimizers:

• RAdam

• Utils:
  • TensorOps:
    • Function to reshape to at least 1d
    • Function to reduce a tensor by a given reduce type
  • Function to recursively load a objects in a config by import paths

Note: So far all the functionality was tested but is not tested with CI/CD and unit tests

All losses and metrics are available in a functional interface and a modular interface.
For all models the dimension parameter can be varied from 1d over 2d to 3d, since von n>3 most operations (including convolutions) are not defined.