It's amazing how much information is conveyed just by a handful of dots in motion. I had a lot of fun a few years ago turning some Michael Jackson moves into an animated constellation: <a href="https://codepen.io/sdellis/full/PGLLwx/" rel="nofollow">https://codepen.io/sdellis/full/PGLLwx/</a>
Link to the demo: <a href="https://storage.googleapis.com/tfjs-models/demos/posenet/camera.html" rel="nofollow">https://storage.googleapis.com/tfjs-models/demos/posenet/cam...</a>
This is great. I have a huge set of professional creative portraits (>1k finished portraits / >3k unfinished) that are all already indexed for poses with a manual data set. Interested in comparing?
This is amazing. I've been wondering of using such technology in training tennis, so that you get a real-time feedback loop on where you should correct your movement. Will definitely explore it more.