Abstract
Accurate eye segmentation can improve eye-gaze estimation and supportinteractive computing based on visual attention; however, existing eyesegmentation methods suffer from issues such as person-dependent accuracy, lackof robustness, and an inability to be run in real-time. Here, we present theRITnet model, which is a deep neural network that combines U-Net and DenseNet.RITnet is under 1 MB and achieves 95.3\% accuracy on the 2019 OpenEDS SemanticSegmentation challenge. Using a GeForce GTX 1080 Ti, RITnet tracks at $>$300Hz, enabling real-time gaze tracking applications. Pre-trained models andsource code are available https://bitbucket.org/eye-ush/ritnet/.
Quick Read (beta)
loading the full paper ...