Hi, I am trying to use the gaze data reading from the gaze CSV files. It seems that 1) some takes have personalized eye gaze and some do not, and 2) the gaze CSV files do not cover all the video frames in some cases. So, what is the difference between the personalized eye gaze and the general eye gaze? Plus, is the eye gaze filed in the same fps as the video? Any information you can give about using the gaze data will be valuable.
Best regards.
I recommend you to read this notebook to how you can use the eye gaze data and other MPS artifacts
- You can read up on Eye Gaze from the “MPS” section of the wiki
- For eye gaze, the above links to Aria’s official documentation
is the eye gaze filed in the same fps as the video?
The eye gaze is determined/computed from the eye tracking image (video) streams (left and right eyes) from the aria device. The frame rate is less than the RGB stream. You should be able to get the frame rate from the projectaria_tools python library (see usage in the above notebook).
So, what is the difference between the personalized eye gaze and the general eye gaze?
To put it simply: personalized eye gaze is from using the eye gaze model that has been calibrated from the calibration step (performed by the participant/camera wearer) prior to recording.
some takes have personalized eye gaze and some do not
Unfortunately not all recordings were calibrated due to a variety of reasons (logistical, technical).
Hi Miguel, thank you for your answer, it is really helpful. Just for the record, if somebody else is interested in the future. The gaze features provided in the CSVs are 10 fps and can be mapped to the frames as coordinates where the gaze is pointing. We could prove that and make use of it in our work.