EyeLoop Playground is a supplementary test space for EyeLoop, a Python 3-based eye-tracker tailored specifically to dynamic, closed-loop experiments on consumer-grade hardware. EyeLoop Playground contains test footage and data to get started.
Click here to access the data.
To achieve good eye-tracking performance, EyeLoop's binarization parameters should be optimized to your video material. Users can adjust two parameters, namely:
Pupil: R/F – Corneal reflection: W/S
Pupil: T/G – Corneal reflection: E/D
When the binary threshold is turned down, dark pixels dominate, which can make the pupil less likely to stand out:
When the binary threshold is too high, light pixels dominate:
The binary threshold should be set somewhere in-between to optimize the pupil contour:
When the Gaussian power is too low, the pupil appears grainy. This can introduce noise, making eye-tracking less ideal:
When the Gaussian power is too high, the pupil might blend into adjacent tissue, making it less discernible:
The Gaussian power should be set to maximize discernibility, while minimizing noise:
When the parameter set has been optimized, EyeLoop will automatically save it for subsequent use. See below on how to load the parameter file.
To load a parameter file, pass it to EyeLoop via command line argument
eyeloop --params [path-to-parameter-file]
On the first run, EyeLoop will automatically calibrate its blink detection to your video footage. This generates a calibration file that can be loaded for subsequent trials.
To load a blink calibration file, pass it to EyeLoop via command line argument
eyeloop --blink [path-to-calibration-file]
Any eye-tracking footage (preferably raw video material) are welcome.
Contact Simon Arvin at [email protected] to contribute.