MinVIS: A Minimal Video Instance Segmentation Framework without Video-based Training

The Y ouTube-VIS 2019/2021 datasets are under CC BY 4.0 License, and Occluded VIS is under CC Tables with standard deviations are shown in Table 6, Table 7, and Table 8 . MinVIS consistently outperforms Mask2Former-VIS in all settings. MinVIS with X% means sub-sampling the annotated frames in training. Our 1% results already outperform previous state-of-the-art. MinVIS significantly outperform existing approaches on OVIS.