Iteratively Selecting an Easy Reference Frame Makes Unsupervised Video Object Segmentation Easier

Unsupervised video object segmentation (UVOS) is a per-pixel binary labeling problem which aims at separating the foreground from background in without using ground truth (GT) mask of object. Most previous UVOS models use first frame or entire as reference to specify Our question why should be selected used mask. We believe that we can select better achieve performance than only frame. In our paper, propose Easy Frame Selector (EFS). The EFS enables us an "easy" makes subsequent VOS become easy, thereby improving performance. Furthermore, new framework named Iterative Mask Prediction (IMP). framework, repeat applying given and selecting "easier" iteration, increasing incrementally. IMP consists EFS, Bi-directional (BMP), Temporal Information Updating (TIU). From proposed state-of-the-art three benchmark sets: DAVIS16, FBMS, SegTrack-V2.