Open-vocabulary video instance segmentation on VEval YT-Temporal-1B (test)
71.2cgF1Human
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Human2025.11 | 71.2 | 97 | 73.2 | 78.4 | 68.8 | 89.7 | 84.3 | |
| SAM 32025.11 | 50.8 | 89 | 57.2 | 69.9 | 60.2 | 81.7 | 70.5 | |
| SAM 3 Detector + T-by-D as tracker2025.11 | 47.6 | 93 | 51.3 | 68.2 | 60.8 | 77 | 70.4 | |
| SAM 3temporal disambiguation=false2025.11 | 47.3 | 84 | 56.3 | 67.8 | 57.1 | 81.2 | 69.8 | |
| LLMDet + SAM 3 TrackerZero-shot=true2025.11 | 8 | 33 | 24.1 | 37.9 | 18.7 | 77 | 33.4 | |
| GLEEPrompting strategy=one NP at a time, Zero-shot=true2025.11 | 2.2 | 22 | 9.9 | 18.9 | 8.5 | 42.9 | 22.5 | |
| GLEEPrompting strategy=all NPs at once, Zero-shot=true2025.11 | 1.6 | 24 | 6.7 | 16.7 | 6.8 | 42.6 | 16.5 |