Video Instance Segmentation on LV-VIS
30.3APGLEE-Plus
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GLEE-PlusBackbone=Swin-Large, Zero-shot evaluation=true, Video-free training=true2023.12 | 30.3 | 31.6 | 29.3 | |
| GLEE-ProBackbone=EVA-02 Large, Zero-shot evaluation=true, Video-free training=true, Inference resolution=high2023.12 | 23.9 | 24.6 | 23.3 | |
| OV2Seg-BBackbone=ViT-B, Zero-shot evaluation=true, Video-free training=true2023.12 | 21.1 | 27.5 | 16.3 | |
| TROY-VISBackbone=EfficientViT-L2, FPS=7.42026.06 | 20.8 | 24.4 | 18.1 | |
| GLEE-LiteBackbone=ResNet-50, Zero-shot evaluation=true, Video-free training=true2023.12 | 19.6 | 22.1 | 17.7 | |
| SegFSSlow Model=TROY-VIS, Backbone=EfficientViT-L2, FPS=23.32026.06 | 18.3 | 21.2 | 16.3 | |
| Detic-XMemZero-shot evaluation=true, Video-free training=true2023.12 | 16.3 | 24.1 | 10.6 | |
| OV2Seg-R50Backbone=ResNet-50, Zero-shot evaluation=true, Video-free training=true2023.12 | 14.2 | 17.2 | 11.9 | |
| Detic-SORTZero-shot evaluation=true, Video-free training=true2023.12 | 12.8 | 21.1 | 6.6 |