The Eye Tribe

If I may ask a quite technical question here,
could you tell me what the unit of PupilSize actually is?

I think the unit of PupilSize is not the same as
that of either PupilCenterCoordinates or RawCoordinates
because I tried to WriteLine the actual values
with the following code in GazeDot::OnGazeUpdate().

Code: Select all

    System.Console.WriteLine("{0} {1} {2} {3} {4}",
        gazeData.LeftEye.PupilSize, gazeData.RightEye.PupilSize,

        //Pupil Distance
        gazeData.RightEye.PupilCenterCoordinates.X - gazeData.LeftEye.PupilCenterCoordinates.X,
        gazeData.RightEye.SmoothedCoordinates.X    - gazeData.LeftEye.SmoothedCoordinates.X,
        gazeData.RightEye.RawCoordinates.X         - gazeData.LeftEye.RawCoordinates.X
        );


An example line of the output is like this.
The two pupil sizes are written in a clearly different scale from the other ones.

> 22.993221282959 22.993221282959 0.169162213802338 6.513671875 -7.025146484375

In GazeEye.cs, I found the following comments,
which probably mean the unit of pupil size is the same as that of pupil center.

Code: Select all

        /// <summary>
        /// Pupil center coordinates in normalized values
        /// </summary>
        [JsonProperty(PropertyName = Protocol.FRAME_PUPIL_CENTER)]
        public Point2D PupilCenterCoordinates { get; set; }

        /// <summary>
        /// Pupil size in normalized value
        /// </summary>
        [JsonProperty(PropertyName = Protocol.FRAME_PUPIL_SIZE)]
        public double PupilSize { get; set; }


On the other hand, I found another description about pupil size
in the tutorial page http://dev.theeyetribe.com/tutorial/.
This means pupil size is represented by the on-screen unit.

... where the OnGazeUpdate is of most interest since it contains the coordinates of
the estimated on-screen gaze position, size of the pupils, position relative to the sensor etc.

Do you know the actual unit of pupil size or how to calibrate it?

Hi Ikeda,

The PupilSize right now is a relative value that is not calibrated to a physical dimension. You can use it to see if the pupil dilates or contracts and by how much compared to the original state.

The PupilCenter is output in normalized coordinates 0 to 1. This is because using sensor pixel values would require knowing resolution of the sensor to compute the location. Normalized coordinates makes it easier to draw a trackbox and mapping coordinates.

We have considered using normalized coordinates for the estimated on-screen gaze coordinates as well but that makes it more difficult for developers since most programming relies on screen pixels. It would be cumbersome to fetch the screen resolution and convert coordinates for every sample. We could of course output both absolute and normalized, would you prefer?

A upcoming release scheduled for February will output pupil size in absolute value, with a millimeter resolution.

Hi,

Thank you for the quick reply. I understand it.

Of course, I'd prefer output both absolute and normalized.
At least, I hope to get some conversion parameters to the PupilCenter's dimension.

ikeda