Video Instance/Panoptic Segmentation

We won the 2nd place in the Pixel-level Video Understanding in the Wild Challenge VPS Track, CVPR 2023. We won the 2nd Place in the 5th Large-Scale Video Object Segmentation Challenge, VIS Track, ICCV 2023.

CTVIS: Consistent Training for Online Video Instance Segmentation. International Conference on Computer Vision, ICCV Paris, 2023. K. Ying, Q. Zhong, W. Mao, Lin Yuanbo Wu, H. Chen, Y. Liu, C. Fan, Y. Zhuge, Chunhua Shen .

Project synopsis: We are dedicated to a stable solution to online VIS that can leverage historical instance embeddings. The other thing more interesting is we creatively implant/infuse noise disturbance into the relevant embeddings around its near intervals, which is demonstrated to be salutary/instrumental to instance discrimination.