You are viewing a single comment's thread from:
RE: AI Camera for Raspberry Pi/Arduino
That's pretty cool, indeed.
Of course, one of the immediate questions is how hard is it to increase the size of the recognized objects from the image feed? 20 is nice, but odds are they aren't the most important 20 to my particular project. Is that something that is trivial or does it require retraining the neural network from more appropriate data?
Posted using Steeve, an AI-powered Steem interface
They have a converter tool that allows you to convert your tf and tf lite models to the format that can be used by KPU. So, the solution here would be to train your NN for recognizing specific objects you need. I plant to publish a tutorial on how to do that as well
https://www.instructables.com/id/Transfer-Learning-With-Sipeed-MaiX-and-Arduino-IDE/
Here is my new article on how to train your custom model! Since you mentioned it before, thought it might be useful.