Add tutorial involving integrating Tiled Python client with Tensorflow ImageDataGenerator
https://www.tensorflow.org/api_docs/python/tf/keras/preprocessing/image/ImageDataGenerator
This is a neat idea. I could also see this being an example notebook.
Is this worthwhile at this point, or do we want to wait until we've have some sort of asynchronous batching mechanism...which would probably be what the keras user would want to be using? https://github.com/bluesky/tiled/issues/125
I would argue we wouldn't have to wait for #125. Demonstrating the integration has value, the overhead of separate requests probably wouldn't be so high with (large) image data, and it would be easy to update a working tutorial to leverage batching once it's available.
With #132 there is now a path to download everything as HDF5, so I definitely think this doesn't need to wait for #125.