How to derive the eye gaze vector from eye position and the eye gaze point of regard?

Hi everyone,

I am using the gaze deepstream example to load and run the gaze model. I get the gaze model outputs on each image which includes the gaze point of regard x, y, z and the gaze vector theta, phi. According to this document, the output x,y, z is the location where the person is looking relative to the camera. And the model page mentions that “The eye gaze vector can also be derived from eye position and eye gaze points of regard.

Can anyone please show me how to convert the gaze point of regard x, y, z to the coordinates relative to the eye positions (e.g. left eye or right eye positions that can be obtained from the facial landmarks model) instead of the camera coordinates?


Let’s discuss in the same topic How to visualise the 3d gaze vector output of the GazeNet model?