Add support/documentation for creating/using custom models

[hmcty]

This library hints at the ability to train and apply custom models, but it's not straightforward.

From the provided train.py script, you do have the ability to train a model but integrating with the rest of the pkg doesn't seem to be supported. The script exports to onnx, which from what I can tell, you can't easily input as a model path to the supported tasks.

If I am missing some documentation or easy way of going doing this, let me know. Otherwise I hope we can consider this as a feature request.

[kimanton]

I have similar question. How to use the provided code to train or fine-tune a model?
I would appreciate some if there were some instructions

Also, when I list available models it shows only these two, are there more available? Or how can I add my own model in the future?
>>> zoo.list_models()
['p2p_surface_3d_DigitSensor.pth', 'touchdetect_resnet_DigitSensor.pth']

[tatsukamijo]

I created a repository for making a custom model: https://github.com/tatsukamijo/DIGIT_train
Hope it helps!