Hi there, I was just curious if anyone has had any success running tensorflow lite on the V5 brain? If so, this could potentially open up a lot of exciting possibilities for somehow hijacking the Vision sensor’s video feed and providing some proper computer vision functionality to a V5 robot. The V5 Brain should be powerful enough, especially seeing as Tensorflow Lite is specifically designed to run on microcontrollers even less powerful than the Brain.
Unless I am mistaken, the vision sensor runs on a closed, isolated system than the V5 Brain. And the only thing the vision outputs is the results of what it sees. That being said, you can’t modify how the vision sensor reads the data, sadly.
Although the vision sensor by itself is not the best, you can make it pretty decent by trying to filter the noise out. Theo has showcased his EMA filter code here for the vision sensor.
Awesome, thanks! That EMA filter thread was really helpful with improving the accuracy of the vision sensor. I guess I now have to figure out how to program it all! Thanks again!