Efficient Pupil Detection with a Convolutional Neural Network

The paper proposes an efficient convolutional neural network for pupil detection capable of state-of-the art accuracy without pretraining. The system architecture consists of twelve convolution layers and two max pool layers with only 108450 prediction parameters. The validation conducted on ExCuSe datasets show that our solution achieves state-of-the-art performance in terms of detection rate within a pixel error bound.