TensorFlow or Caffe

I am a novice in Deep Learning (although I have some experience in Data Analysis and programming) and I am starting a project to do image recognition. My plan is to train some models in our GPU server and then deploy them in Tk1 and/or Tx1.
My plan was to use Caffe as it seems to integrate well with the Jetson. But with the recent release of TensorFlow I wondered whether Caffe is still the right choice…
Any insights?

Looks like Caffe is more complete and faster now (at least on x86, don’t know about ARM), but TensorFlow is newer and apparently may be the “wave of the future”. Here’s a discussion on the pros/cons of each: https://groups.google.com/forum/#!topic/caffe-users/1rfDUZFlMgQ