Deep learning algorithms and applications

Abstract

This paper presents an appearance-based gaze-tracking implementation called Browser Eye Tracker (BET). BET is a convolutional neural network for real-time (>30fps) eye- tracking that can run on any device with a web browser without first downloading anything or buying specialised eye-tracking webcams. BET achieves a prediction error 2% lower than previous in-browser approaches on average. An in-browser Auto Sampler (AS) for automated sample collection, a Gaze-tracking Playground (GP) for comparing different models and Real-Time Prediction Testing (RTPT) were also implemented as part of the project.