Hacker News Reader

YOLO, YOLO2 and YOLO9000 are pretty amazing pieces of work. Resnet variations gets more attention and wins more competitions, but in a lot of spaces YOLO* is much more useful.

It deals with all the practical issues of making image recognition fast and scaleable by making sensible tradeoffs.

The youtube video demonstration of video image recognition on YOLOv2's website is entertaining as well..

https://pjreddie.com/darknet/yolo/

In that video when there's a shot of Judy Dench, YOLO recognises "Bench". So close, yet so far.

Very fun! I'll have to spin this up and play around with it. Hopefully, I'll have 1.1.0 TensorFlow/Raspberry Pi binaries compiled and available by the end of the weekend.

One of the side-goals I've been meaning to make time for is training (or porting) a smaller CNN targeted to run closer to real time on the Raspberry Pi- I think a project like this would benefit from that.

You might be interested in SqueezeNet - https://github.com/DeepScale/SqueezeNet, I imagine this would run quite fast even on the Raspberry Pi.

BerryNet – Deep Learning Gateway on Raspberry Pi