Added
- download scripts for example data
- model zoo repo
- simple super-resolution network
- full example training and deploy scripts on medical images for regression, classification, segmentation, representation learning, super-resolution w/ downloadable pre-trained models.
- continuous testing, code linter and automatic doc generation
Changed
- unet implementation now has convolutional blocks symmetrically in the encoder and decoder
- core modularisation based on AbstractModule replaced by tf.layer and tf.estimator
- core io AbstractReader class now wraps tf.contrib.data.Dataset, replacing v0.1 reader and queuing classes
- tutorials are now changed to explain how dltk modules can be used with tf
- improved readme in terms of contributing, installation instructions
- switched to google docstrings and loose coding style
Removed
- sonnet like modularisation: AbstractModule classes
- summary modules, instead relying on problem-specific tf.summaries in estimator model_fns
- examples based on natural images (MNIST, CIFAR10, etc.)