Caffe2 is a deep learning framework made with expression, speed, and modularity in mind. It is an experimental refactoring of Caffe, and allows a more flexible way to organize computation.
Caffe2 started as an attempt to improve the design of caffe, because we have learned quite a lot since the 2 years of running and using caffe. It is also aiming to make Caffe more general, not only focusing on vision but applicable to all kinds of machine learning tasks.
Some highlights of the Caffe refactoring idea can also be found at the slides from the CVPR caffe tutorial.
Note that this is not meant for a complete replacement for Caffe right now. You are more than welcome to try it and we appreciate much if you could contribute back, but a lot of things may still be missing (e.g. CuDNN) so expect a few bumps as an early adopter. Caffe is still stably supported and for a production environment you may want to continue using it.
I (Yangqing) would also like to personally thank Google, my current employer, to allow me to continue devoting my personal time to work on open-source projects like Caffe.
Caffe2 is released under the BSD 2-Clause license.