Add keras-based models, add pixel-wise loss, other improvements

• d47eae91 Add dataset_to_hdf5 command
• cc8a142a EPSC estimators and losses
• 8e274774 improve block extraction
• b9583f15 Add a parallel read option to dataset_from_tfrecords
• a8bc6541 Add an extra feed option to the extractors
• 1c3fdb10 improve logging for vat loss
• 9058f1a8 Add more logging to bob tf cache-dataset
• e7445a23 Save one best evaluation checkpoint by default
• c82154ef Allow architectures to provide the logits layer
• 281e1c26 Add more logging to the predict commands
• ac7d3eda Add a pixel wise loss
• 2f7fec82 Add keras-based models
• c9311f9c Move utils to the utils folder
• 70e0b851 Allow shuffle on epoch end in generator
• 54a80323 add a function to load png images
• 036a308f improve logging
• e155ed50 improve logging in losses
• fa765388 add center loss, mmd loss, and pairwise confusion loss
• d957c74a make sure densenet layer names are consistent
• 63993d46 nitpick
• cda150df add starting point options to the style transfer script
• 7fabc04e Add more utils
• cb9744bc new keras models
• 38de7bdc Add GAN tools
• f6e1bb57 implement random rotate
• aece6e1b improve the generator and biogenerator classes
• 6e6addea nitpick
• b9cc5a9a handle nans in network predictions
• 27e73aec cast labels to required format
• 8082804e fix imports
• b9fd76bd Add model_summary to keras utils
• 4a714e00 add more option to reproducible
• 3241c38a Add keras version of models
• f481311f remove six dependency
• 36c3f4b9 Add a Gaussian blur filter