Brief

Researchers have developed EyeTheia, an open-source, lightweight deep learning pipeline for gaze estimation using standard webcams. The system combines landmark extraction with a convolutional neural network, offering real-time tracking suitable for browser-based experiments and clinical research. Performance evaluations on the MPIIFaceGaze dataset showed that both adapting pre-trained models and training from scratch yielded comparable results before calibration, with user-specific fine-tuning further reducing prediction errors. EyeTheia demonstrated strong agreement with a commercial tracker in a Dot-Probe task, providing a transparent and extensible solution for low-cost gaze tracking. AI