In the task of human segmentation, there were more than a dozen of teams. Using deep CNNs, we won the first place with a large margin. The average overlapping rate between our predictions and the ground-truths is 86.83%.