Preprocessing on PoG in GazeCapture and MPIIFaceGaze

[AbdirayimovS]

Dear author,

Thanks for great contribution. I am wondering regarding preprocessing you mentioned in the paper: Section IV.Experiments: "We use physical distance to the camera in both training and testing for its device independence"

• GazeCapture provides PoG distance to the camera in centimeters. It is written in next paragraph: "The dataset provides physical screen sizes and 2D gaze position in pixels. So we converte pixel coordinates of the gaze to physical coordinates relative to the screen."

My question: As far as I understood, gazecapture's and MPIIFaceGaze's PoGs are in physical coordinates, but they have different origin. (For gazecapture, it is camera location, while for mpiiface, it is top left corner.) Is it correct how I understood your point?

My thinking: You have converted MPIIFaceGaze from pixels coordinates tot he physical coordinate relative to the camera. (Assuming camera in laptops are located in middle of x-axis).

Looking forward to hearing from you!

Sincerely, Sardor