This document describes a study that used capsule networks to estimate gaze from eye images. The researchers developed a two-step approach called Gaze-Net that first classifies gaze direction and then estimates gaze. Gaze-Net was trained on a large dataset and achieved good accuracy. It was also able to be personalized to new users through transfer learning, improving performance on a separate dataset. The study demonstrated that ocular images contain sufficient information for decoding head pose and eye orientation to estimate gaze direction.