Last week, in my post, I’ve shared with you my experiments using the Movidius Neural Compute Stick and the Raspberry Pi. This week, I’ve continued my experimentation switching to video streams (from still images), using a new ANN, and putting at work multiple NC. Both series of experimentation turned out to be successful. On the highlight, the ANN has an improved success rate in its recognition and the Pi handles the video steam @ ~50% CPU utilization.
In particular, my beloved kitties are consistently recognized as such and not as birds :-). Not too shabby! On the lowlight, however, I am still failing to build TensorFlow – few of its dependencies in fact – for/on the Pi. Also, when running the two NCSes alongside the USB camera (not the camera module), the Pi is really unstable and freezes (and/or reboots) regularly. To be continued…
k
I like it