Zhengkai Jiang, Zhangxuan Gu, Jinlong Peng, Hang Zhou, Liang Liu, Yabiao Wang, Ying Tai, Chengjie Wang, Liqing Zhang
Video Instance Segmentation (VIS) is a task that simultaneously requires classification, segmentation, and instance association in a video. Recent VIS approaches rely on sophisticated pipelines to achieve this goal, including RoI-related operations or 3D convolutions. In contrast, we present a simple and efficient single-stage VIS framework based on the instance segmentation method CondInst by adding an extra tracking head. To improve instance association accuracy, a novel bi-directional spatio-temporal contrastive learning strategy for tracking embedding across frames is proposed. Moreover, an instance-wise temporal consistency scheme is utilized to produce temporally coherent results. Experiments conducted on the YouTube-VIS-2019, YouTube-VIS-2021, and OVIS-2021 datasets validate the effectiveness and efficiency of the proposed method. We hope the proposed framework can serve as a simple and strong alternative for many other instance-level video association tasks.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Video Instance Segmentation | YouTube-VIS validation | AP50 | 57.2 | STC (ResNet-50) |
| Video Instance Segmentation | YouTube-VIS validation | AP75 | 38.6 | STC (ResNet-50) |
| Video Instance Segmentation | YouTube-VIS validation | AR1 | 36.9 | STC (ResNet-50) |
| Video Instance Segmentation | YouTube-VIS validation | AR10 | 44.5 | STC (ResNet-50) |
| Video Instance Segmentation | YouTube-VIS validation | mask AP | 36.7 | STC (ResNet-50) |
| Video Instance Segmentation | OVIS validation | AP50 | 33.5 | STC (ResNet-50) |
| Video Instance Segmentation | OVIS validation | AP75 | 13.4 | STC (ResNet-50) |
| Video Instance Segmentation | OVIS validation | mask AP | 15.5 | STC (ResNet-50) |