KirstensGitHub
commented
5 years ago Updates for selecting gaze data: compile_gaze_data_sustained (3).pdf
First, we see that when data is states other than state 7, it only records gaze data at the single point (0,0); as such, we eliminate gaze data collected when the tracker is not in state 7. Then, we verify that the bottom left corner (0,0) is the only corner the data indicates subjects stare exactly, directly at with both eyes over the course of the experiment(s). As such, we feel confident that we can further remove any additional (0,0) datapoints.
Then, we explore the average distance between average gazepoints reported for the left and right eyes. We find that ~80% of the datapoints show the gaze location of the left eye to be 5cm or less away from the location of the right eye along both the horizontal and vertical axes ( a max possible diagonal distance of ~7 cm ). Additionally, we find that the mean of all the gaze coordinates for datapoints within this range fall within the bounds of our screen (near fixation) but that the mean for those outside of this range yield an x value well outside the bounds of our screen (~270 cm to the right of our screen's edge). As such, we exclude datapoints where it is reported that the two eyes are gazing more than 5cm apart in the horizontal or vertical directions.
Finally, we average over the x- and y-coordinates for the two eyes, and remove any datapoints falling outside the bounds of our screen.
Here, we review our gaze data collection methods and analysis code. You can see the code for all checks / images below, in the following PDF's: Gaze_Check.pdf, compile_gaze_data_sustained (3).pdf
CHECKS
CHECK 1: First, we show that the EyeTribe is recording gaze data from the full screen (despite our screen being larger than 24").
CHECK 2: Second, we explore how many gazepoints, from the times when composite images are on screen, fall just outside the screen edges.
Experiment 1:
Experiment 2:
CHECK 3: Third, we ensure that the gaze data is being properly subselected for each run and trial, based on time stamps in the csv.
Experiment 1:
Experiment 2:
CHANGES & CONCERNS: Code was converting timestamps in the gaze data using a package that truncates the times at tenths of a second; updated to convert timestamps with millisecond precision.
Did a more thorough diagnosis of aberrant gaze datapoints and updated code for more uniform elimination of bad data. (See explanation and PDF of jupyter notebook in a comment below.)
Reading over the documentation and reviewing the reduced screen size respresented in the Eyetribe GUI (.4 size of our screen), the resolution of our gaze data might be lower than the standard eyetribe resolution.