What is Point Supervision Worth in Video Instance Segmentation?

Video instance segmentation (VIS) is a challenging vision task that aims todetect, segment, and track objects in videos. Conventional VIS methods rely ondensely-annotated object masks which are expensive. We reduce the humanannotations to only one point for each object in a video frame during training,and obtain high-quality mask predictions close to fully supervised models. Ourproposed training method consists of a class-agnostic proposal generationmodule to provide rich negative samples and a spatio-temporal point-basedmatcher to match the object queries with the provided point annotations.Comprehensive experiments on three VIS benchmarks demonstrate competitiveperformance of the proposed framework, nearly matching fully supervisedmethods.