Automatic optic disk (OD) segmentation is an important tool for early detection of eye diseases. In this article, we proposed a Res-UNet network by applying residual learning module and other improvements in U-Net for optic disk segmentation in retinal image. Since training data available is insufficient, we enlarge the data set by generating data pieces. Res-UNet is then trained to classify each pixel of the input retinal image. Finally, the predicted probability map is further post-processed with morphological technique to get final OD segmentation result. Experiments on the public DRISHTI-GS data set including comparison with the best known methods show that the proposed model outperforms most existing methods on several metrics.