I appreciate the printed direction under the video of my face, but I feel like it might have been a little bit easier for me to learn to direct it if there was also a face mesh drawn on the displayed camera frame as debug info. This has seemed to allow me to adjust easier to facial controls in the past, and other people I polled about the helpfulness of displaying such debug info to the end user seemed to overwhelmingly agree (polls were of about 20 people, and super unscientific -- this is essentially still anecdotal advice to be taken with a grain of salt, but I got literally no negative reactions).
As an example of how this could help learning, I jumped straight to the demo without really reading the instructions (hurpadurpa), and I foolishly assumed that you were just detecting head position in the frame and using the coordinates, rather than the direction my face was pointing, to determine control direction. It wasn't until I looked at the code that I realised what I was doing wrong. I probably could have read more and picked up on this without diving into the code, but I also probably would have picked it up immediately if I saw a mesh of some facial landmarks (rather than assuming an invisible bounding box). I think smaller adjustments would be assisted as well -- the user can figure out how far off screen their face can go while still being a face way easier if you show them what you think looks like a face.
Another maybe-helpful-maybe-not thought: adding an additional direction indicator to the art in the main game screen might make it easier to get used to the controls as well, and let the user know exactly when they trigger a change in direction without drawing their eyes away from the action. This could be as simple as two pixels for eyes on the snake head.
Very cool, but that video works by using normal supervised learning (to detect hand gestures) while mine detects the direction your face is pointing in using pose estimation.
How hard would it be to use these tools to detect someone touching their face and throwing a chime to help remind them not to? It could run in the background all day for laptop/webcam users. #covid etc.
Very impressive! Callibration took a fair minute or two but I didn't have to do anything complicated like show it all the different poses.
Fun couple of games. I had trouble getting it to recognise down but I think that's because I can't bend my neck forward and keep looking at the screen (glasses) to see the snake moving.
Also I am definitely team: Screen edge should wrap around!
