search

emilianavt / OpenSeeFace

Robust realtime face and facial landmark tracking on CPU with Unity integration
BSD 2-Clause "Simplified" License
1.41k stars 151 forks source link

gaze estimation output format #2

Closed hwuebben closed 4 years ago

hwuebben commented 4 years ago

Hey there,

First of all thanks for the awesome work!

I`m trying to grab the gaze estimation results but do not understand the format of the data. I concluded that the second and third (indices 1 and 2) entries of the eye_state variable show the current gaze, but what exactly do the numbers mean? Is the gaze relative to the head position or the camera? Also what is the immediate output of the gaze model?

Can you maybe give me some clarification?

Thank you and all the best!

hwuebben commented 4 years ago

Ah so I guess I misunderstood the gaze estimation concept, the model does not provide a gaze vector but the exact position of the pupils' centers which I then can use to determine the gaze, right?

emilianavt commented 4 years ago

Hi and thank you!

Yes, that is correct. You can find the way I calculate the gaze directions here: https://github.com/emilianavt/OpenSeeFace/blob/master/Unity/OpenSee.cs#L177

I basically assume an average human eyeball size and calculate the direction from its center to the pupil center.

Also what is the immediate output of the gaze model?

It's a heatmap with with offsets. It works basically the same as the face landmarks, but only for a single landmark per eye.

emilianavt commented 4 years ago

This seems to be solved, so I'll close this issue. If you have any other questions, please feel free to reopen it.