Eye tracking device calibration

A method for pre-calibrating an eye tracking apparatus of a field-ready head-mounted computing device, the method comprising:

• generating a generic training data set that comprises a first image of a trainer's eye and a first image label that comprises a first designated location on a generic calibration image displayed on a first graphical user interface output for display by a data-collection head-mounted computing device being worn by the trainer, and wherein the first image is captured by a first image sensor on the data-collection head-mounted computing device at a moment in time when the trainer is gazing at the first designated location, wherein the generic training data set comprises data instances generated by multiple trainers having a diversity of eye characteristics;
• generating a trained machine learning model using the generic training data set as a training input by using a regression process whereby the trained machine learning model is able to output a gaze location given an unlabeled eye image when trained;
• saving the trained machine learning model on the field-ready head-mounted computing device;
• generating a user-specific training data set that comprises a second image of a user's eye and a second image label that comprises a second designated location on a user-specific calibration image displayed on a second graphical user interface output for display by the field-ready head-mounted computing device being worn by the user, and wherein the second image is captured when the user is gazing at the second designated location on the user-specific calibration image, wherein the user-specific calibration image has less designated locations than the generic calibration image;
• generating a user-specific correction factor for the trained machine learning model using the user-specific training data set;
• capturing an image of the user's eye by an image sensor on the field-ready head-mounted computing device;
• and determining, by the trained machine learning model and using the user-specific correction factor, a first gaze direction for the user's eye using the image of the user's eye as input.

One or more computer storage media that, when executed by a computing device, causes the computing device to perform a method for pre-calibrating an eye tracking apparatus of a field-ready head-mounted computing device, the method comprising:

• generating a generic training data set that comprises a first image of a trainer's right eye, a second image of the trainer's left eye, and a first image label that comprises a designated location on a calibration image displayed on a graphical user interface output for display by a data-collection head-mounted computing device being worn by the trainer, and wherein the first image is captured by a first image sensor on the data-collection head-mounted computing device at a moment in time when the trainer is gazing at the designated location, and wherein the second image is captured by a second image sensor on the data-collection head-mounted computing device at the moment in time when the trainer is gazing at the designated location, wherein the generic training data set comprises data instances generated by multiple trainers having a diversity of eye characteristics;
• generating a trained machine learning model using the generic training data set as a training input by using a regression process whereby the trained machine learning model is able to output a gaze location given unlabeled eye images when trained;
• saving the trained machine learning model on the field-ready head-mounted computing device;
• capturing a first unlabeled image of the user's right eye by a first image sensor of the field-ready head-mounted computing device;
• capturing a second unlabeled image of the user's right eye by a second image sensor of the field-ready head-mounted computing device;
• and determining, by the trained machine learning model, a first gaze direction for the user's eye using the first unlabeled image and the second unlabeled image as input.