The eyevec-control application enables you to perform a gaze error analysis for all (i.e. many) combinations of calibration pattern and gaze mapping polynomial type. This allows you to find out which calibration pattern and polynomial type seems most suited (at least for the participants you run this test on).
The general procedure is as follows:
-
Perform a calibration with many points; preferably 9 points or more.
-
In the Test & Record page:
-
select a dot pattern type with many points; preferably more than for calibration
-
tick the Analyse gaze error checkbox
-
run the test
-
-
Visit the analysis directory and inspect the latest two files created:
<yyyymmdd>T<hhmmss>-<refnum>.txt(human readable output)
and
<yyyymmdd>T<hhmmss>-<refnum>.csv(data, CSV)
Generally both files generated will be very large. To somewhat restrict
the size of the .txt file results with a too high error will have been
excluded, nevertheless the file will still be inconveniently long.
|
Here’s the output of one gaze error analysis run:
20251219T152221-0.txt
20251219T152221-0.csv
Combinations tested
Which calibration types (patterns) will be evaluated depends on the calibration you ran. The possible calibration types are given by the EyeVecCalibrationType enumeration. All calibration types except the custom ones define targets on a 5x5 grid. The calibration types evaluated are all those that represent a subpattern of the calibration pattern chosen (the 1-dot pattern doesn’t count). So 5-square covers 4-square as well as the 3 and 2 slope patterns, but not 4-diamond etc. For the gaze error analysis test it is best to always chose the 9-point calibration type or higher.
Which polynomial types will be evaluated depends on the number of targets in the calibration (sub)pattern evaluated. Only well-determined and overdetermined systems can be evaluated. This means a 5-coefficient polynomial type will only be used for (sub)patterns with at least 5 points. The possible polynomial types are given by the EyeVecPolynomialType enumeration.