Night Eyes: A Reproducible Framework for Constellation-Based Corneal Reflection Matching

This release provides automatically generated identity-preserving corneal reflection (glint) annotations for OpenEDS 2019 [1] and OpenEDS 2020 Sparse Segmentation [2] Dataset. The annotations were produced using the Night Eyes constellation-based matching framework and extend OpenEDS beyond pupil segmentation by introducing structured multi-glint correspondence labels aligned with the fixed LED layouts of the respective datasets. For each frame, detected glints are assigned consistent LED identities according to a predefined constellation template. The annotations make it possible to evaluate multi-glint detection and identity-preserving correspondence algorithms under realistic recording conditions, including occlusion, spurious reflections, and illumination variability. The current version contains the raw generated annotations without manual correction. While the underlying pipeline has been systematically evaluated, residual correspondence or localization errors may remain. A reviewed and refined version of the annotations is planned in a future update. The goal of this release is to support reproducible benchmarking and structured evaluation of geometry-based and learning-based glint matching approaches. These annotations are derived from OpenEDS and are released under the same CC BY-NC 4.0 license as the original dataset. Sources [1] Garbin, S. J., Shen, Y., Schuetz, I., Cavin, R., Hughes, G., & Talathi, S. S. (2019). Openeds: Open eye dataset. arXiv preprint arXiv:1905.03702. [2] Palmero, C., Sharma, A., Behrendt, K., Krishnakumar, K., Komogortsev, O. V., & Talathi, S. S. (2020). Openeds2020: Open eyes dataset. arXiv preprint arXiv:2005.03876.