Hi author here. We just pushed the code on github yesterday, I did not expect this to end up on HN so early. I will add the relevant information about which frameworks we are using to the README.

However, it is much more than a Keras wrapper. It supports models based on any framework through a plugin system. We plan to add support for pytorch and others.

Ok, cool. Why would I want to use your library, and not, say, Tensorpack, which, in addition to everything you promise, provides dozens of high quality model implementations, and is very fast (i.e. much faster then Keras)?

Thanks for mentioning tensorpack, looks very interesting!

VergeML is not a library, it is a (command line) tool which lets you drop images in a directory and type 'ml train' in the most basic case.

You would use VergeML to quickly train with everyday datasets.

VergeML is framework agnostic, there will be model implementations in pure TensorFlow ("closer to the metal") and other libraries. Keras is a very good fit for the classifier based on transfer learning from ImageNet. It provides a lot of functionality and good training performance.