So far I have had no luck using retrain.py to make a graph that that mvccompile works with.
So if I wanted to use my own custom data set and derive my own inferences what do I do so I can get
the maximum acceleration out the Nerual compute stick.
I have seen YOLO work at about 10fps on a PI but that is with canned video and not a live source.
Best I have been able to achieve is about ~3fps.
That all being said this is useless without being about to make/modify a graph from scratch.
With version 2 of the software is there youtube video or anything that can teach me how to get
this thing to work. If it takes 2 sticks, (one for object detection and one for inference detection) I have 2 sticks.
Please check this tutorial:
We have succeeded custom training it. There are some tweaks to be considered. Look at my comments in the same article