I was wondering the same thing. I kept using Caffe because when I did some tests in early 2017, it performed best on inference performance on Jetson TX1s from all networks tested. However earlier this year I found some bugs in the reshape layer implementation (which is kinda important when dealing with LSTMs)
which I tried to fix:
and there was simply no response at all for 3 months, which is kinda alarming considering the severity of the bug (at least for training)
I'm now in the process of porting all my stuff to pytorch in the hope I can then use caffe2 for inference.
But a life sign from the dev team would still be nice. At least an official end of life announcement or something.