gazelib-fixtures

This package contains 14 human gaze recordings to be used for unit testing and algorithm design.

Abbreviations:

• SRT = Saccadic Reaction Time
• SD = Saccade Duration
• C = Center of the screen
• UR = Upper Right corner
• LR = Lower Right corner
• UL = Upper Left corner
• LL = Lower Left corner

Two (2) fixations

We asked the participant to stare at C, recorded the gaze position on screen with an eye tracker, and then selected two representative sets of 1000 sample points.

Samples in fixtures/fixation-gazepoints.*.json:

• [0]: fixation of 3.3 sec without blinks
  • 1000 sample points
  • 300 Hz sampling rate
  • 0 invalid sample points
• [1]: fixation of 3.3 sec with 3 blinks
  • 1000 sample points
  • 300 Hz sampling rate
  • 102 invalid sample points

Twelve (12) saccades

Each saccade sample contains 300 sample points recorded at 300 Hz. At the first sample point a stimulus image was revealed at one of the four corners. After SRT sample points the gaze starts to move and the movement lasts for SD sample points.

Samples in fixtures/trials-gazepoints.*.json:

• [0]: direct saccade from C to UR
• [1]: direct saccade from C to LR
• [2]: direct saccade from C to LL
• [3]: direct saccade from C to UL
• [4]: multipart saccade from C to UR
• [5]: multipart saccade from C to LR
• [6]: multipart saccade from C to LL
• [7]: multipart saccade from C to UL
• [8]: curved multipart saccade from C to UR
• [9]: curved multipart saccade from C to LR
• [10]: curved multipart saccade from C to LL
• [11]: curved multipart saccade from C to UL

We extracted SRT and SD from a video recording (120 Hz) for each trial. They are available in fixtures/trials-features.json.

File formats

The samples come in two formats: pointlist and standard.

*.pointlist.json

A pointlist file is a list of samples. Each sample is a list of points. Each point is a list [x, y]. The x and y are combined average coordinates of both eyes.

Invalid points equal to [null, null].

*.standard.json

A standard file is a list of samples. Each sample is a list of points. Each point is a dictionary with the following keys:

• left_pupil_diam: The diameter of left pupil. Value can be null. Example value: 3.190
• left_x: The horizontal gaze position of the left eye. Relative to screen size. Value can be null. Example value: 0.507
• left_y: The vertical gaze position of the left eye. Relative to screen size. Value can be null. Example value: 0.541
• right_pupil_diam: The diameter of right pupil. Value can be null. Example value: 3.190
• right_x: The horizontal gaze position of the right eye. Relative to screen size. Value can be null. Example value: 0.515
• right_y: Vertical gaze position of the right eye. Value can be null. Example value: 0.549
• time: Sample capture time in microseconds since UNIX epoch. Value cannot be null. Example value: 1448373364122185

Data collection method

The data has been collected through the following experiment. The participant is asked to follow a white square on a screen. For the first 10 seconds, the square stays still at the center of the screen. The participant stares at the square and blinks his eyes a few times. After the 10 seconds, 12 attention shifting trials were played. With each trial, the square first appears at the middle, and after two seconds, moves to a corner. For the first 4 trials, the participant was asked to follow the square as quickly as possible. For the next 4 trials, the participant was asked to pause the saccade between the center and the corner and then continue to the corner. For the last 4 trials, the participant was asked to do arc-like saccades.

Tobii TX300 eye tracker was used to capture the data.

Detailed description of the used collection methods can be found at docs/method-description.pdf.

Original recordings

The original recordings are contained in four files:

• original-recordings/recording-001-fixation.gazedata
• original-recordings/recording-001-trials.gazedata
• original-recordings/recording-001-video-annotations.json
• original-recordings/recording-001-video-diff-enhanced.mp4

The gazedata files are tab-separated-values files with following columns:

• LeftEyePosition3dRelativeX
• LeftEyePosition3dRelativeY
• LeftEyePosition3dRelativeZ
• LeftEyePosition3dX
• LeftEyePosition3dY
• LeftEyePosition3dZ
• XGazePosLeftEye: Relative horizontal position on the screen
• YGazePosLeftEye: Relative vertical position on the screen
• LeftEyePupilDiameter
• ValidityLeftEye: Tobii validity for eye recognition, 0 = perfect, 1 = good
• RightEyePosition3dRelativeX
• RightEyePosition3dRelative
• RightEyePosition3dRelativeZ
• RightEyePosition3dX
• RightEyePosition3dY
• RightEyePosition3dZ
• XGazePosRightEye
• YGazePosRightEye
• RightEyePupilDiameter
• ValidityRightEye
• TETTime: UNIX timestamp in microseconds
• stim: ID of stimulus
• aoi: Location ID of stimulus
• tag: Phase of trial, Wait = image at center, Target = image at corner
• trialnumber: Sequence number of trial, starts from 0
• starttime: Phase start UNIX timestamp in microseconds
• aoi_coord: Relative coordinates of stimulus location
• endtime: Phase end UNIX timestamp in microseconds

Developer notes

The video files are large. GitHub supports large files if Git Large File Storage is used. Therefore, before $ git push, install Git Large File Storage (git-lfs).

License

MIT