Video Instance Segmentation

Video instance segmentation is a new task in the field of computer vision that aims to simultaneously achieve instance detection, segmentation, and tracking in videos. This task extends the problem of image instance segmentation to the video domain for the first time, promoting the development of related research through large-scale benchmark datasets such as YouTube-VIS, which includes 2,883 high-resolution videos, 40 categories of labels, and 131,000 high-quality instance masks, making it highly valuable for practical applications.