Abstract

From scientific research to commercial applications, eyetracking is an important tool across many domains. Despiteits range of applications, eye tracking has yet to become apervasive technology. We believe that we can put the powerof eye tracking in everyone’s palm by building eye trackingsoftware that works on commodity hardware such as mobilephones and tablets, without the need for additional sensors or devices. We tackle this problem by introducing GazeCap-ture, the first large-scale dataset for eye tracking, contain-ing data from over 1450 people consisting of almost 2.5Mframes. Using GazeCapture, we train iTracker, a convolu-tional neural network for eye tracking, which achieves a sig-nificant reduction in error over previous approaches while running in real time (10–15fps) on a modern mobile device. Our model achieves a prediction error of 1.71cm and 2.53cm without calibration on mobile phones and tablets respectively. With calibration, this is reduced to 1.34cm and2.12cm. Further, we demonstrate that the features learnedby iTracker generalize well to other datasets, achieving state-of-the-art results. The code, data, and models are available at http://gazecapture.csail.mit.edu.