I'm no expert in anything deep learning, but I think getting it to work on mobile is just a matter of importing your trained file like the dlib facial landmark detector. which definitely runs in real time - albeit maybe not at 60 FPS.
yup, I actually use the dlib facial detector in this app for a ventriloquist type feature (not shown in video). The facial landmarks detector assume you've framed an ROI for a face already. Locating faces in an image is trivial... locating hands in an image and their orientation, not so much : ) That's not to say a CNN couldn't be trained to do it, it's just going to take some work!
5 comments
[ 4.2 ms ] story [ 27.0 ms ] threadand also
> hand gesture recognition is far more complicated than facial feature detection
the good news is we may not be that far off anymore https://news.ycombinator.com/item?id=14263678