Tensorflow - YOLO Or Similar For Object Classification And Recognition

While ARC's object recognition is good, Tensorflow's is better. It's better because it is pretrained with a large library (200?) of common objects. Pre-trained object classification software saves one the work of feeding the camera 5000 photos of dogs or 2000 photos of cups or 9000 photos of people, etc. Would ARC work with 2000 photos of a dog?

I see that there are stand-alone boards such as the HuskeyLens that could be interfaced to ARC with a lot less trouble, but to have recognition/classification in software would be the ultimate goal. So, if not Tensorflow (or similar) then how about a hardware interface to boards such as HuskeyLens?