Re:Deep belief image recognition on Raspberry Pi
by bigiain on Monday January 25, 2016 @01:02AM
(#10965837)
Apart from a recent (26 day old) update to the javascript, this looks like it got abandoned back when Google acquired Jetpac (who's website now seems defunct). Anyone know what the status of this is? I guess there must've been some more-recent-than_aug-2014 updates, since it mentions RaspberryPi2s...
Not really RPi related - but I think it's only a matter of time until we have mobile GPU optimised BLAS libraries working with Mali, Adreno etc. At that point machine learning will become more feasible on current mobile devices
Re:Deep belief image recognition on Raspberry Pi
by buserror on Monday January 25, 2016 @08:06AM
(#10966925)
Had had a good play with that lib earlier on debian, it's pretty cool! Very lightweight (considering) and I tossed it a few pictures of stuff, and it's quite impressive. I particularly liked my eBay shopping history. Didn't like my landscape photos so much, but it's likely to be that it hasn't been trained on any...
What would be fantastic is a way to recover some sort of rough coordinates of what the classification have found.
http://petewarden.com/2015/05/18/jetpacs-deep-learning-frame...
https://github.com/tensorflow/tensorflow/issues/445#event-48...
https://github.com/teradeep/demo-apps/tree/master/generic-em...
or this:
http://mxnet.readthedocs.org/en/latest/tutorial/smart_device...
Not really RPi related - but I think it's only a matter of time until we have mobile GPU optimised BLAS libraries working with Mali, Adreno etc. At that point machine learning will become more feasible on current mobile devices
What would be fantastic is a way to recover some sort of rough coordinates of what the classification have found.
I had trouble getting Caffe and Torch to run natively (i.e. MS toolchain), anyone knows of a good Windows port for these that one can link to?