GazeTrack

𝗚𝗮𝘇𝗲𝗧𝗿𝗮𝗰𝗸 is a 𝗳𝘂𝗹𝗹-𝘀𝘁𝗮𝗰𝗸 𝘄𝗲𝗯 𝗮𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻 that enables 𝗰𝗹𝗶𝗲𝗻𝘁-𝘀𝗶𝗱𝗲 𝗲𝘆𝗲 𝘁𝗿𝗮𝗰𝗸𝗶𝗻𝗴 for video-based user studies. I 𝗹𝗲𝗱 𝘁𝗵𝗲 𝘁𝗲𝗮𝗺 and 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗱 𝘁𝗵𝗲 𝗠𝗟 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲 𝘂𝘀𝗶𝗻𝗴 𝗧𝗲𝗻𝘀𝗼𝗿𝗙𝗹𝗼𝘄.𝗷𝘀 for in-browser face and eye detection, training a lightweight 𝗰𝘂𝘀𝘁𝗼𝗺 𝗖𝗡𝗡 𝗼𝗻 𝗮 𝗰𝘂𝗿𝗮𝘁𝗲𝗱 𝗱𝗮𝘁𝗮𝘀𝗲𝘁. The model is 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗲𝗱 𝗶𝗻 𝗿𝗲𝗮𝗹-𝘁𝗶𝗺𝗲 through user calibration to improve gaze accuracy (~80%). I also 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗲𝗱 𝗵𝗲𝗮𝘁𝗺𝗮𝗽 𝘃𝗶𝘀𝘂𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻𝘀 over video content using 𝗠𝗮𝘁𝗽𝗹𝗼𝘁𝗹𝗶𝗯 and 𝗙𝗙𝗺𝗽𝗲𝗴, allowing researchers to easily analyze gaze distribution. The platform is fully containerized using 𝗗𝗼𝗰𝗸𝗲𝗿 for straightforward deployment.