A pupillary image dataset: 10,000 annotated and 258,790 non-annotated images of patients with glaucoma, diabetes, and subjects influenced by alcohol, coupled with a segmentation performance evaluation

The Pupillary Light Reflex (PLR) is the involuntary movement of the pupil adapting to lighting conditions. The measurement and qualification of this information have a broad impact in different fields. Thanks to technological advancements and algorithms, obtaining accurate and non-invasive records of pupillary movements is now possible, expanding practical applications. Visual attention tracking enables the development of solutions for Eye Tracking Marketing or Eye Gaze Marketing, optimized gaming interactions, drowsiness detection in drivers, and, more recently, diagnostic support applications. These advancements have been made possible by algorithms and publicly available datasets to improve these algorithms. However, it is important to note that most of these datasets only provide information from healthy individuals. This article introduces and publicly shares a diverse dataset with three distinct subsets: recordings of individuals who underwent supervised alcohol consumption, individuals diagnosed with type II diabetes mellitus, and individuals diagnosed with glaucoma in early, moderate, and severe stages. In addition to the data, to assist researchers aiming to conduct studies involving pupillary behavior, the study evaluates pupillary segmentation and eye-tracking algorithms, highlighting the superior accuracy of YOLOv7 in calculating pupillary diameter compared to classical approaches. By utilizing the proposed dataset, the research advances the field of pupilometry-based diagnostics, promoting reliability and effectiveness and indicating the most precise methods for pupillary segmentation. •Eye images of glaucoma (early, moderate, severe), diabetes, and alcohol users.•10,000 images with manual annotations for training ML models or neural networks.•Dataset of 258,790 eye images from the mentioned classes.•Comparing classical algorithms and CNNs for pupil segmentation.