Just spent the afternoon working with ML5 with ZIM. Goodness, it was so easy. Why did we not do this years ago! I kept telling people, if you want hand tracking, face tracking, body tracking, etc. use TensorFlow. ML5 is based on TensorFlow and makes it easier. That is what P5js is using. But, I never went there myself to see.
Will take a look at bringing the ml5 into perhaps the cam module so we can just import cam.
Strange - it does for us - try just the URL do you see it?
We added a callback (modelLoaded) to the first picture one - maybe it was needed? Not sure, like I said, worked for us without it but on the cam ones, it did not work unless we added the callback. So try again, and if anyone has problems, let us know.
Hmmm... ml is 4mb minified so perhaps we will not automatically add it with the cam module. There is probably an es6 module or just use it as a script tag.
Found there is a general JavaScript videoWidth and videoHeight issue where we are getting 640x480 even though our camera is 1280x720. If we specify like this:
const cam = new Cam({
width:1280,
height:720,
config:{
width: 1280, // otherwise, seems to get the video as 640x480
height: 720
}
});
Then it seems to work - but not sure what happens on a 640x480 cam. Ran into a couple cases with different computers not doing it right. So will continue to test but for now am hardcoding the 1280x720
Ml5 is so cool
bodypose works with sticky man dancing befor cam
how cool
the easy bodypose skeleton
what do you think..
