An Efficient Dense Network for Semantic Segmentation of Eyes Images Captured with Virtual Reality Lens

Eye-tracking and Gaze estimation are difficult tasks that may be used for several applications including human-computer interfaces, salience detection and Virtual reality amongst others. This paper presents a segmentation algorithm based on deep learning that efficiently discriminates pupils, iris, and sclera from the background in images captured using a Virtual Reality lens. A light network called DensetNet10 trained from scratch is proposed. It contains fewer parameters than traditional architectures based on DenseNet which allows it to be used in mobile device applications. Experiments show that this network achieved higher IOU rates when comparing with DensetNet56-67-103 and DeeplabV3+ models in the Facebook database. Furthermore, this method reached 8th place in The Facebook semantic segmentation challenge with 0.94293 mean IOU and 202.084 parameters with a final score of 0.97147.