I'm an intern at Numenta this summer and I just started working on cleaning up 
some of the old vision code in NuPIC. The ImageSensor region currently found in 
NuPIC hasn't been changed in quite some time so unfortunately Subutai's 
assessment of using the current implementation of NuPIC for image problems is 
still accurate. This is not to say that NuPIC can't be used for image problems, 
but rather we don't have any specialized vision implementation built on top of 
NuPIC's current algorithms.

If you're still interested in experimenting with vision using NuPIC, the MNIST 
example in nupic.research may be a good jumping off point 
(https://github.com/numenta/nupic.research/tree/master/image_test). It uses the 
older ImageSensor code but might offer some inspiration as to how to begin with 
vision experiments.

Jared

From: nupic 
<[email protected]<mailto:[email protected]>> on 
behalf of miguel angel rodriguez 
<[email protected]<mailto:[email protected]>>
Reply-To: "NuPIC general mailing list." 
<[email protected]<mailto:[email protected]>>
Date: Tuesday, June 30, 2015 at 12:25 PM
To: Jared Weiss <[email protected]<mailto:[email protected]>>
Subject: Re: NuPIC for image recognition

I would certainly try, although I'm working in the Sound field so no experience 
there ;)

2015-06-30 16:28 GMT+02:00 Ricardo Franco 
<[email protected]<mailto:[email protected]>>:
Some time ago I remember Subutai told that the current implementation of NuPIC 
wouldn't have good performance in Image Recognition problems.

With the recent updates, this situation changed? The current implementation is 
suitable for image problems?

Thanks.

--

Ricardo Franco Andrade

Web Developer

email: [email protected]<mailto:[email protected]>
skype: ricardo.krieg
phone: +55 (86) 9569 8521<tel:%2B55%20%2886%29%209569%208521>
linkedin: http://br.linkedin.com/in/ricardokrieg/
github: https://github.com/ricardokrieg



Reply via email to