VOGeo-Gaze: Calibration-Free, Geometry-Aware Deep Learning for Real-Time Gaze Tracking in Clinical Video-Oculography

Quantitative eye movement analysis is important for neurological diagnostics, yet existing video-oculography (VOG) systems typically require calibration, device-specific settings, or accurate gaze labels. We present VOGeo-Gaze, a real-time, calibration-free, geometryaware neural network that estimates gaze by reconstructing anatomically meaningful eyeball parameters from image features. The method combines segmentation-driven projection geometry, a refraction-aware pupil correction module, and temporal anatomical stabilization, so gaze is derived from interpretable eye geometry rather than direct angular regression. Trained only on the public TEyeD dataset with weak gaze supervision, VOGeo-Gaze was evaluated on 116 clinical recordings from 17 patients and 19 healthy subjects using EyeSeeCam, a clinical gold-standard VOG system. It achieved median absolute angular errors of 0.33° horizontally and 0.35° vertically, with nearly 92% of recordings below 1° error while operating at >300 FPS. These results demonstrate sub-degree clinical gaze estimation without subject-specific calibration, camera intrinsics, or accurate gaze labels, providing a scalable and interpretable alternative to conventional VOG pipelines. Code is available at https://github.com/DSGZ-MotionLab/VOGeo-Gaze .