Robust Video Object Segmentation with Restricted Attention

This paper focuses on the two problems of the similar objects distraction and the lack of robustness for unseen object categories in semi-supervised video object segmentation task. Existing methods have achieved great results on the benchmark dataset, but these two problems still have not been completely solved. We propose the Robust Video Object Segmentation With Restricted Attention (RVOSR), which can suppress the effects caused by similar objects and filter out noise confusion from other irrelevant regions. Meanwhile augmenting the semantic information of features, which makes the features more suitable for video object segmentation task. Extensive experiments demonstrate the effectiveness of our approach and achieve the state-of-the-art performance on the widely-used VOS benchmarks including DAVIS-2016 (92.1% $\mathcal{J}{{\& }}\mathcal{F}$), DAVIS-2017 (86.8% $\mathcal{J}{{\& }}\mathcal{F}$) and YouTubeVOS-2019 (84.8%).