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

Virmarie Maquiling; Yasmeen Abdrabou; Enkelejda Kasneci

cs.CVApr 2, 2026

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

Authors: Virmarie Maquiling, Yasmeen Abdrabou, Enkelejda Kasneci

Organizations: Human-Centered Technologies for Learning, Technical University of Munich Munich, Germany · Munich Center for Machine Learning (MCML) Munich, Germany

Abstract

Corneal reflection (glint) detection plays an important role in pupil-corneal reflection (P-CR) eye tracking, but in practice it is often handled as heuristics embedded within larger systems, making reproducibility difficult across hardware setups. We introduce a 2D geometry-driven, constellation-based pipeline for mulit-glint detection and matching, focusing on reproducibility and clear evaluation. Inspired by lost-in-space star identification, we treat glints as structured constellations rather than independent blobs. We propose a Similarity-Layout Alignment (SLA) procedure which adapts constellation matching to the specific constraints of multi-LED eye tracking. The framework brings together controlled over-detection, adaptive candidate fallback, appearance-aware scoring, and optional semantic layout priors while keeping detection and correspondence explicitly separated. Evaluated on a public multi-LED dataset, the system provides stable identity-preserving correspondence under noisy conditions. We release code, presets, and evaluation scripts to enable transparent replication, comparison, and dataset annotation.

Figures & tables

Figure 1. Overview of the proposed multi-glint correspondence pipeline. The input image undergoes enhancement, over-detection with appearance scoring and adaptive fallback, candidate pooling and gating, and finally template-based matching using a predefined LED constellation. The output assigns identity-consistent glints to the detected reflections.

Figure 2. Per-subject results from the winning hyperparameter sweep evaluated on Chugh et al. ’s dataset ( 2021 )

Figure 3. Success and failure cases across three eye trackers with different LED configurations. Images are from Chugh et al. ’s dataset ( Chugh et al., 2021 ) , OpenEDS2019 ( Garbin et al., 2019 ) , and OpenEDS2020 ( Palmero et al., 2020 ) . Blue circles are the detected glints while the green circles show template-projected positions based on the chosen candidate glints.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Sweep stage	Best run ID	Matcher	Acc. $\uparrow$	Prec. $\uparrow$	ID-free Acc. $\uparrow$	Med. Err. (px) $\downarrow$
Tier A (coarse)	tierA/run_008	hybrid	0.7289	0.8082	0.7787	1.416
Tier B1	b1_005	sla	0.7319	0.7956	0.7760	1.423
Tier B2 (winner)	tierB/b2_081	sla	0.7414	0.8116	0.7740	1.414
Tier B3	tierB/b3_004	sla	0.7381	0.8042	0.7680	1.416

Appendix

Table 1. Best-performing configuration from each sweep stage.

Run ID	Ties	Acc.	Prec.	IDF	Med.Err.	FB
tierB/b2_081	8	0.7414	0.8116	0.774	1.414	99,98,97 / 4 / 8
tierB/b2_113	8	0.7364	0.8081	0.767	1.415	99.5,99,98.5,98 / 4 / 12
tierB/b2_065	8	0.7346	0.8147	0.766	1.414	99,98,97 / 2 / 8

Appendix

Table 2. Collapsed Tier B2 sweep summary with fallback settings.

Glint	Present	Pred.	Correct	Acc. $\uparrow$	Prec. $\uparrow$	Mean Err. (px)	Med. Err. (px)
G0	1522	1485	1127	0.740	0.759	10.324	1.568
G1	1845	1623	1382	0.749	0.852	8.060	1.350
G2	1789	1576	1319	0.737	0.837	10.699	1.302
G3	1737	1553	1256	0.723	0.809	11.003	1.407
G4	1870	1768	1413	0.756	0.799	11.726	1.477

Appendix

Table 3. Per-glint performance for the frozen Tier B2 configuration.

Eye Tracking

Eye tracking measures where people look and how their eyes move to study visual attention, behavior, and gaze-controlled systems.

See all 61 papers

Seeing Through the Glare: A Multi-Source Benchmark and Ocular-Adaptive Pixel MeanFlow for Eyeglass Reflection Removal
Oct 7, 2026Tao Liu, Youwei Pang, Kailai Zhou +7Image Restoration
AI4X Team, Nanyang Technological University, Singapore · X3000 Inspection Co., Ltd. · Tsinghua University, China +2
An Affordable, Wearable Stereo-Eye-Tracking Platform
Apr 27, 2026Alexander Zimmer, Yasmeen Abdrabou, Enkelejda KasneciWearable Sensing Eye Tracking
Human-Centered Technologies for Learning, Technical University of Munich, Munich, Germany
Lazy or Efficient? Towards Accessible Eye-Tracking Event Detection Using LLMs
Apr 14, 2026Dongyang Guo, Yasmeen Abdrabou, Enkelejda KasneciEye Tracking
Technical University of Munich, Munich, Germany · Munich Center for Machine Learning (MCML), Munich, Germany · Munich Center for Machine Learning (MCML), Germany

Oct 7, 2026J/K move · Enter open · S save

Seeing Through the Glare: A Multi-Source Benchmark and Ocular-Adaptive Pixel MeanFlow for Eyeglass Reflection Removal

Tao Liu, Youwei Pang, Kailai Zhou, Jiaming Zuo+6

AI4X Team, Nanyang Technological University, Singapore · X3000 Inspection Co., Ltd. · Tsinghua University, China+2

Image Restoration

Open paper AskPDF arXiv

Eyeglass reflection removal is important across smartphone imaging, video conferencing, and other face-centric visual applications. The task is challenging because reflections range from mild photometric contamination to severe ocular occlusion, requiring selective correction and plausible reconstruction without altering identity or natural appearance. Existing datasets cover limited reflection conditions, constraining generalization to complex real-world scenes and systematic evaluation. We introduce OcuBench, a multi-source benchmark comprising 10,280 controllable synthetic pairs, 732 real-input pseudo-pairs, and 458 independent real-world test images, supporting both paired evaluation and assessment beyond generated supervision. We further propose OcuFlow, an ocular-adaptive pixel MeanFlow (pMF) framework for efficient, detail-preserving restoration. It combines geometry-adaptive representation with one-step pMF to focus reconstruction on reflection-obscured ocular regions, together with native-resolution frequency-preserving synthesis to retain reliable observed details. Experiments across diverse reflection conditions demonstrate that OcuFlow achieves consistent advantages in reflection removal quality, ocular fidelity, and efficiency. In a blind user study, it receives

67.32\%

of selections,

6.2\times

the next-best share. Both the code and dataset will be released.

Abstract

Eye Tracking

Explore similar work

Seeing Through the Glare: A Multi-Source Benchmark and Ocular-Adaptive Pixel MeanFlow for Eyeglass Reflection Removal

An Affordable, Wearable Stereo-Eye-Tracking Platform

Lazy or Efficient? Towards Accessible Eye-Tracking Event Detection Using LLMs