glassesValidator v1.2.5

• Now use median instead of mean when computing offset (acc) data quality measures. Since a fixation classifier is used that is robust to spikes in the data (I2MC), median is a better representation of the offset of the fixation intervals delivered by I2MC to the validation targets
• Fix two crashes in msgbox that pops up when an error occurs
• SeeTrue STONE importer:
  • ensure frames are in the right order also on file systems that don't provide alphabetically ordered file listings
  • removed two checks that the manufacturer said are not needed
• Hopefully make precompiled exe able to find the include ffmpeg.exe

glassesValidator v1.2.4

• Renamed SeeTrue to SeeTrue STONE. This is a breaking change in that projects with data from the SeeTrue in them from versions prior to 1.2.4 cannot be opened in this version anymore. You have to remove the recordings using the older version and reimport them using this version.
• Added hardcoded default camera calibration matrix for SeeTrue STONE

glassesValidator v1.2.3

Fix regression where glassesValidator didn't work when you do not have a camera calibration and therefore only have gaze on plane by means of homography.

glassesValidator v1.2.2

• deploy config was not updated for new OpenCV version

glassesValidator v1.2.1

• fixed building of exe, again

glassesValidator v1.2.0

• Improved logic for determining timestamps of frames in mp4 files. Now takes edit lists (thanks Tim Cornelissen) and composition offset atoms into account. This improves sync of data to video on multiple devices, notably the Tobii Glasses 3.
• Can now correctly handle video files with local corruption (gaps) in them, also such as produced by the Tobii Glasses 3 on occasion.
• Added an option (default on) to the logic for matching fixations to validation targets that removes any overall shift in the data before performing the matching. This allows glassesValidator to handle recordings where data shows large overall shifts: assignments of fixation points to validation targets are now correct in that case.
• Added an option to limit the maximum distance of a fixation point to a validation target for it to be assigned to a validation target. This is implemented as a factor (default 0.5) of the minimum inter target distance of the validation targets.

NB: for fixation matching to return to the same behavior as for version 1.1 and older, switch off the option to remove the overall shift called Use global shift in the GUI) and set the maximum distance factor (Matching distance factor in the GUI) to a large value.

glassesValidator v1.1.3

Now catch case where a recording does not have dataquality, and provide an understandable error message when attempting to export

glassesValidator v1.1.2

Update imgui_bundle dependency to 0.9.0, and pin to this specific version as future versions may not be backward compatible

glassesValidator v1.1.1

• Prevent crash in detect_markers() and make_video() when none of the detected ArUco markers are part of the poster
• compute_offsets_to_targets(): don't crash but provide useful error when there is no data for a marker interval
• detect_markers() and make_video(): deal with broken video (OpenCV's video reader returning false for a frame before the video is supposed to have ended): don't stop but continue processing to next frames.

glassesValidator v1.1.0

• Add support for AdHawk MindLink
• Add support for new data quality type, pose_world_eye, see Readme
• re-add information about number of ArUco markers used for pose estimation (was removed with port to OpenCV 4.8)
• port visualization in utils.makeVideo() to imgui based gui, so it works again when headless opencv is installed