"Near-eye image"

"TEyeD is the largest public near-eye image dataset captured by head-mounted eye trackers, including VR\/AR devices. It contains over 20 million real-world images with rich annotations: 2D\/3D landmarks, pupil\/iris\/eyelid segmentation, 3D eyeball models, gaze vectors, and eye movement types. Covering driving, simulation, indoor and outdoor scenarios, it includes reflections, shadows, and occlusions to simulate real degradations. TEyeD provides a reliable benchmark for eye segmentation, gaze estimation, and sensor-based human\u2013computer interaction research in VR\/AR systems. Due to computational constraints and research requirements, a representative subset of 3,200 samples was selected for this study, which includes: 1,700 clear eye images, 500 images with specular reflections, 500 images under non-uniform illumination, and 500 images with partially occlusion."