Hello guys !
Like some of you know, I worked during +10 years in Deep Learning (at the time, so called “AI” was barely known) and I’m passionnate into this field
After the last meeting at FrameVR with @gabrieljbaker I started to do some tests with some face detection stuff.
I open this topic to discuss / show / have fun, about some tests, running deep NNs directly inside the browser, in JS.
As a start, here is a face landmark model (WASM + a bit of GPU acceleration) taken from mediapipe, and running directly in the web page, alongside BabylonJS
Mouth :
Eyes :
Headpose :
Face Crop Tracking :
NB: Please note that I don’t do any image cropping here. The plane texture still receive the full image camera, I only tweak the texture UVs in realtime 
Next steps (soon)
- Gaze (eye direction)
- Animation an avatar with it
@gabrieljbaker I think this kind of techs could leverage the usage of FrameVR when used from Browser 
?
