This is the best kind of pointless/awesome. How was the learning experience with the PoseNet ML model? Did you have any experience with machine learning or image processing beforehand? Seems like a great way to learn about both.
Thanks! I had played with ML before, and did some more traditional style computer vision in the past (with morphological transforms and stuff in OpenCV), but I had no idea that the browser was capable of such feats — I was really impressed that PoseNet was able to run fast enough for me to call it every frame.
Honestly, there were many opportunities for learning that I didn’t take, simply because I wanted to get a prototype up and running asap. For example, I was really interested in learning the linear algebra behind the perspective transform (for turning the webcam’s coordinates into the game’s coordinates) and the math behind detecting collisions (like the separating axis theorem), but I opted to npm install those instead.
Maybe I will rewrite those parts myself next weekend :)