Minor adjustment in dependency versions.

v2.0.1

v2.0.1: Adjust some versions and docs.

March 2020 Maintenance: improve docs, fix a bug and bump dependencies.

Changelog

• CheckpointManager API has completely changed and usage has been updated.
• Update Config docs: use literalincludedirective to save duplication in docs.
• Bugfix: Explicitly cast image features to float in Dataset classes.
• Remove redundant dependencies.
• Bumpy PyTorch, torchvision and h5py versions.

Pretrained checkpoints are available with v1.0.0 tag, and can be used here.

Stable release - code for our ICML 2019 paper.

With pre-trained checkpoints on all training phases, with config files used to train them.