Neon Accuracy Test Report

We present the results of a validation study, assessing the gaze-estimation accuracy of the Neon eye-tracking system by Pupil Labs. We provide results obtained from, 1) a screen-based experiment conducted in a controlled lab environment, and 2) from ‘in-the-wild’ scenarios. In particular, we show that - without a user-specific calibration - Neon achieves a per-subject median accuracy of 1.8° at 1.3m depth and across a 60° x 35° field of view. By applying a simple offset correction, median per-subject accuracy can be further improved to 1.3°. Furthermore, our results highlight that Neon operates robustly in lighting conditions ranging from total darkness to bright sunlight, for a representative spectrum of eye appearances, as well as for realistic variations in position on the wearer's head. For instance, we show that median accuracy stays within 0.2°-0.4° between indoor and outdoor settings.

Changes for Version v2:

• The report now contains an additional figure (new Figure 7), which is based on the same data as in Version v1. It is showing the distribution of the median gaze-estimation accuracy over the field of view for the 5% of subjects associated with most accurate and least accurate gaze-estimation performance, respectively, selected based on mean uncorrected accuracy.

• Updated incorrectly stated field of view of the Neon scene camera (correct is 103° × 77°).