Graph Neural Networks for Video Object Segmentation
摘要
As one of the key methods of scene understanding, video object segmentation can be used in applications such as video conferencing, video editing and autonomous driving. Recently, many video object segmentation methods have been proposed. In this chapter, we extend graph neural networks to video object segmentation to enhance the utilization of target structure information. We describe in detail the video object segmentation methods based on graph neural networks in terms of graph construction, positional encoding for graph structure, and graph convolution operations.