This week, I practiced using Teachable Machine image classifier with ml5. I'm interested in finding out what's the best practice of using TM image classifier.
I tested with a simple game, pong. I trained many things from drawn signs to pose. I discovered that TM image classifier works the best when you train an object. The drawn signs sometimes work but sometimes don't. I think it might be problems with the thickness of the lines and lighting. Poses work the worst. It was wrong every time, and it was slow. So, I decided to use objects as left and right. It's not intuitive to play with objects, but the direction is correct and responsive every time in different settings.
Comments