Hi, I have a project running with the face-landmarks-detection from model in version v0.0.3 and I have already done some work to upgrade it to version v1.0.2.

I have been guided a bit by your code since you use the new version.

The problem I have is how they calculate and use the iris points in the normalisation and mapping of the cursor coordinates to the screen space.

Do you have any idea how to deal with this problem? I have the project working with v0.0.3.

Thanks a lot

I went to test out the demo on my iPhone 15 to do a very crude benchmark of performance, but it looks like it is broken on mobile. I tried both sides of the camera front and back, neither work.

While this works great on PC, I think it'll be a bit slow on some low end devices. Are you familiar with react native and would you recommend using tensorflow.js at all for a mobile.