This repository contains our PyGaze code for eyetracking people translating input sentences with the help of images.
The resulting dataset has been released here: github.com/ufal/eyetracked-multi-modal-translation
The experiment can be run in the demo mode with the option --dummy (experiment.py --dummy)
In the dummy mode, when you are presented with the calibration screen, please place your mouse pointer on the black dot to complete the calibration.
The experiment should then start presenting you with the other instructions.
All the test observations are now kept under the folder recordings/.
To test if the web-cam tracking works, run web_cam.py.
To run the experiment with the "webcam-eyetracker", please edit the "webcam_mode.sh" to change your path names and then execute it.
Wait till the window with the output from the tracker loads (could take some time) and then start with the real experiment (entering name of the subject and list number).
Record the audio separately and then add the audio to the recording folder (or modify the audio_module.py file as per your system config).
Also, the output from the web-cam tracker (cam_record_default) gets saved into the experiment/ folder (just like it happens with the actual eye-tracker. Manually move it into the appropriate recordings folder.)
To test if the DSLR mode works, run the test_webcam_dslr.sh script.
If there are errors pertaining to the Liveview mode, change the mode of your camera to any of the P,S,A,M modes.
For a Nikon-3500, the mode that worked was the M (manual) mode.
Adjust the focus on your camera beforehand as close-ups focusing on the pupil tend to get blurry with incorrect focus settings.
To see the output of the DSLR mode, edit the web_cam.py to change the capture device.
The devices are listed under /dev/video*.
You can then run webcam_mode.sh to run the experiment using the DSLR.