Scalable Video Object Segmentation with Identification Mechanism