You could feed in grey scale image values using a scalar encoder to build your SDRs for you, if all you wanted to start with was a single layer that understood the overall scene to some extent.
Though what could be simpler and more useful is ff you have something that will munch on the image before hand, you could have it generate a csv with classifiers for pixel groups and their pixel location in a bound x,y and use a classification encoder for the pixel group classification field, and scalar encoders for the locality values. You can start getting even funkier by tagging an average rgb value or some such field. ie: picture_element, center_pos_x, center_pos_y, outer_bound_x, outer_bound_y eye, 200, 100, 40, 40 ... On Tue, Jul 16, 2013 at 3:49 PM, Sai Kumar Dwivedi <[email protected]>wrote: > Hello Byrne, > > Thank you so much for your detailed explanation. > > Like you said, suppose we give predigested data (face is w in width, h in > height; nose length is n1, colour y3 etc...), then how do we generate the > SDR?? Is there any specific algorithm for encoding the input into Spatial > Pooler or is it just the ON and OFF set of attribute we define. > > Ex- If someone has blue eyes then the input bit is 1 or else 0. > Kindly explain the encoding of input into SDR... > > Sai Kumar Dwivedi > Undergraduate CSE (2nd Year) > NIT Rorkela, India > > > > On Wed, Jul 17, 2013 at 12:39 AM, Sai Kumar Dwivedi > <[email protected]>wrote: > >> Hello NuPIC Community, >> >> Suppose I want implement HTM-CLA in facial recognition. By the previous >> post I learned that generating SDR manually is not an easy task as every >> feature cant be extracted. Then how is the SDR generated ?? Can someone >> explain correlating to facial recognition? >> >> -- >> Sai Kumar Dwivedi >> >> > > > -- > Sai Kumar Dwivedi > Computer Science and Engineering (2nd Year) > NIT, Rourkela > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > >
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
