-
Wear the glasses and open GazeTracker by executing "./opengazer" or opengazer.cbp
-
Make sure the eye image is in the middle of video and the pupil(pink circle) is well detected
-
If not, in MainGazeTracker.cpp line 142 & 143 by changing: Eye_params.Radius_Min & Eye_params.Radius_Max
-
In the UI, click the Calibrate button to start calibration
-
Keep you head stable and follow the points which is shown on the screen
-
The initual points number are 49 you might change it in calibrator.cpp from line 149 to line 197
-
There are 4 choices of the calibration points number which is 13,16,25 & 49
-
If the points are changed, correspondingly, you need to change the calpoints.txt
-
I have modified another two files whcih are calpoints13.txt &calpoints25.txt if you need to use 13 points
calibration. Then just change the calpoints13.txt to calpoints.txt -
After calibration, the blue point will be display on the screen, it is the predicted points based on your eye image
-
Press test button in the UI, it will start testing the accuracy
-
You can change the testing points number on the screen in MainGazeTracker.cpp line 263 & 264 by changing the for loop
-
For the evaluation, in OutputMethod.cpp from line 47 to line 52, it is going to ouput the predict coordinate to file "svm testing".
-
There is an "Evaluation.py" file in the project. Run the file in terminal, type "python Evaluation.py"
-
Further instructions will be in the Evaluation.py.
-
The Gaussian evaluation figure will be shown & you can consider the peak of the curve is average pixel error
boosting / gazetracker Goto Github PK
View Code? Open in Web Editor NEWThis project forked from lefteye47/gazetracker
License: GNU General Public License v2.0