Tackling Background Distraction in Video Object Segmentation

Semi-supervised video object segmentation (VOS) aims to densely track certain designated objects in videos. One of the main challenges this task is existence background distractors that appear similar target objects. We propose three novel strategies suppress such distractors: 1) a spatio-temporally diversified template construction scheme obtain generalized properties objects; 2) learnable distance-scoring function exclude spatially-distant by exploiting temporal consistency between two consecutive frames; 3) swap-and-attach augmentation force each have unique features providing training samples containing entangled On all public benchmark datasets, our model achieves comparable performance contemporary state-of-the-art approaches, even with real-time performance. Qualitative results also demonstrate superiority approach over existing methods. believe will be widely used for future VOS research. Code and models are available at https://github.com/suhwan-cho/TBD .