There is a newer version of the record available.

Published January 30, 2026 | Version 1.0.0
Dataset Open

PopEYE - Infrared Ocular Image Dataset for Eye State and Gaze-Direction Classification

Authors/Creators

  • 1. University of Modena and Reggio Emilia: Modena, IT
  • 2. ROR icon University of Modena and Reggio Emilia

Description

PopEYE is a specialized dataset of 14,976 near-infrared (NIR) ocular images designed to support the development and benchmarking of computer vision algorithms for ophthalmic applications. The dataset focuses on two primary tasks: eye-state detection (open vs. closed) and coarse gaze-direction classification.

Context and Motivation

In clinical ophthalmic examinations (such as Pupillary Light Reflex measurement or Optical Coherence Tomography), patient cooperation and correct eye positioning are critical for data integrity. PopEYE_2022 was developed to train machine learning models capable of real-time monitoring of the eye status, ensuring that only valid frames are processed and providing immediate feedback on patient alignment.

Technical Specifications

  • Imaging Modality: Near-Infrared (NIR) imaging.

  • Optical Setup: Captured using a custom-built Maxwellian-view ophthalmic stimulator, as described in Gibertoni et al. (SPIE 2022).

  • Image Resolution: 772 × 520 pixels (grayscale, PNG format).

  • Dataset Size: ~3.66 GB.

  • Participants: Data collected from 22 different subjects across multiple acquisition sessions to ensure variability in iris patterns, eyelid shapes, and eyelash occlusions.

Dataset Structure and Classes

The images are organized into six mutually exclusive classes based on the eye's state and positioning:

  1. Correct (8,160 images): Eye open, centered, and correctly positioned for measurement.

  2. Closed (1,790 images): Full eye closure (blinks or sustained closure).

  3. Up (1,379 images): Gaze directed upwards.

  4. Down (1,015 images): Gaze directed downwards.

  5. Left (1,296 images): Gaze directed leftward.

  6. Right (1,336 images): Gaze directed rightward.

Key Challenges for AI Models

The dataset intentionally includes common NIR artifacts to test model robustness, such as:

  • Specular reflections: Bright spots from NIR LED sources.

  • Partial occlusions: Eyelids and eyelashes obscuring the pupil/limbus boundary.

  • Anatomical variability: Differences in eye shape and iris pigmentation under NIR.

Related Publications

This dataset has been utilized and validated in the following research works:

  • Sensors 2023 (Vol. 23, Issue 1, 386): Comparative analysis of ML, DL, and Expert Systems for eye classification.

  • SPIE Ophthalmic Technologies XXXV (2025): Real-time monitoring using SVM-based architectures.

Abstract (English)

The PopEYE dataset is a specialized collection of 14,976 near-infrared (NIR) images of the human eye region, specifically designed to support the development and benchmarking of computer vision algorithms for eye-state detection and coarse gaze-direction classification. Each image is provided in a fixed resolution of 772 × 520 pixels in 8-bit grayscale PNG format. The acquisition was performed frontally using a custom-developed Maxwellian-view optical configuration, comprising a board-level CMOS camera and a specialized lens system in which the subject's eye is precisely positioned at the focal point. This setup ensures a high-contrast representation of the anterior segment, making the pupil, iris, limbus, and portions of the sclera and eyelids clearly distinguishable under stable 850 nm infrared illumination.

The dataset is categorized into six mutually exclusive classes identified through manual annotation supported by fixed visual aids and an expert system algorithm. The classification includes a correct positioning class for eyes open and properly aligned for clinical measurements (8,160 images), a closed class representing full eye closures such as blinks or sustained lid closure (1,790 images), and four directional classes representing gaze shifts relative to the central optical axis, specifically up (1,379 images), down (1,015 images), left (1,296 images), and right (1,336 images). The data capture the natural anatomical variability of 22 subjects and incorporate common real-world artifacts, such as specular reflections from NIR sources and partial pupil occlusions by eyelashes or eyelids. By providing standardized labels and high-resolution NIR imagery, PopEYE serves as a robust resource for training machine learning models intended for real-time patient monitoring during ophthalmic examinations.

Files

dataset_report.txt

Files (3.0 GB)

Name Size
md5:4ca79a618f5d38e6a321b7d1ab799abb
895.2 kB Download
md5:f8c490c5ecb1c318c8a24b35c086b1d1
1.3 kB Preview Download
md5:84993b218b8abcabffffd1b2de6b1d6a
493 Bytes Download
md5:81463f4d20dfdd5a38ca0a377c5ff82a
487.6 kB Preview Download
md5:26ba4318b05fbc855d09e448f5f46dce
3.0 GB Preview Download
md5:9db4c0d397afb7c1268201951ad19b7d
4.5 MB Preview Download
md5:dc3273b7b06a44a00e153a7de541dae0
5.7 kB Preview Download

Additional details

Related works

Is described by
Journal article: 10.3390/s23010386 (DOI)
Conference paper: 10.1117/12.3044486 (DOI)
Conference paper: 10.1117/12.2607380 (DOI)

Software

Programming language
Python
Development Status
Active

References

  • Gibertoni, G., Borghi, G., & Rovati, L. (2022). Vision-Based Eye Image Classification for Ophthalmic Measurement Systems. *Sensors*, 23(1), 386. https://doi.org/10.3390/s23010386
  • Gibertoni, G., & Rovati, L. (2025). Enhancing ophthalmic examinations with real-time eye position monitoring using SVM. *Proc. SPIE 13300, Ophthalmic Technologies XXXV*, 1330009. https://doi.org/10.1117/12.3044486
  • Gibertoni, G., Di Pinto, V., Cattini, S., Tramarin, F., Geiser, M., & Rovati, L. (2022). A simple Maxwellian optical system to investigate the photoreceptors contribution to pupillary light reflex. *Proc. SPIE 11941, Ophthalmic Technologies XXXII*, 1194116. https://doi.org/10.1117/12.2607380