Documentation updates and minor bugfixes

From History:

• Deep supervision added
• Documentation updated
• Model IO is now standardized

Multiple new features for training

From History:

• New Network: Light-UNet
• Misc bugfixes
• Automatic check-pointing of the model has been added
• Extending the codebase has been simplified
• New optimizers added
• New metrics added
• Affine augmentation can now be significantly fine-tuned
• Update logic for penalty calculation
• RGB-specific augmentation added
• Cropping added

New metric calculation mechanism

Changelog:

• Misc bugfixes for segmentation and classification
• DFU 2021 parameter file added
• Added SDNet for supervised learning - https://doi.org/10.1016/j.media.2019.101535
• Added option to re-orient all images to canonical
• Preprocessing and augmentation made into separate submodules

Updated feature set for segmentation and classification

Changelog:

• Half-time epoch loss and metric output added for increased information
• Gradient clipping added
• Per-epoch details in validation output added
• Different types of normalization layer options added
• Hausdorff as a validation metric has been added
• New option to save preprocessed data before the training starts

Major refactor finished

• First level of code re-organization
• No API changes for config files

Final tag before major refactor

0.0.8

Merge pull request #39 from sarthakpati/master