500,000 Images Closer to Eyelid and Pupil Segmentation

Human gaze behavior is not the only important aspect about eye tracking. The eyelids reveal additional important information; such as fatigue as well as the pupil size holds indications of the workload. The current state-of-the-art datasets focus on challenges in pupil center detection, whereas other aspects, such as the lid closure and pupil size, are neglected. Therefore, we propose a fully convolutional neural network for pupil and eyelid segmentation as well as eyelid landmark and pupil ellipsis regression. The network is jointly trained using the Log loss for segmentation and L1 loss for landmark and ellipsis regression. The application of the proposed network is the offline processing and creation of datasets. Which can be used to train resource-saving and real-time machine learning algorithms such as random forests. In addition, we will provide the worlds largest eye images dataset with more than 500,000 images DOWNLOAD.