@derekstearns4034
Kevin, this is cool! I am in the process of building my InMoov. This is the sort of program I have been chasing since the Predator TLD algorithm came out 7 years ago. I think when you talk about having secondary neural networks for sub-classification, it will be easier to apply context to the learning and identification, great idea. Specifically, in this case, you have trained the network based on one "sense" alone: sight. But if you were to feed your model data including geolocation, time of day, via other "primary networks" (or maybe using a REALLY BIG convolution network) I think the recognition would get better and be able to achieve a higher library of objects/people! I chose to build my InMoov specifically to try to achieve working with more "senses". I hope to have mine built soon so I can start working with it in the real world.
@matsonnerby
Hi Kevin. This is rellay awsome. Greatings from Mott's
@travisandrew5878
Do you have more information and/or blog on doing this? We love to get someting like this running on our (still in progress) Inmoov
@CyberSyntek