Open-vocabulary video instance segmentation on VEval SA-V (val)
53.9cgF1Human
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Human2025.11 | 53.9 | 86 | 62.4 | 71.7 | 56.9 | 90.8 | 68.9 | |
| SAM 32025.11 | 29.3 | 66 | 44.5 | 60.7 | 44.7 | 83.2 | 57.3 | |
| SAM 3temporal disambiguation=false2025.11 | 24.4 | 56 | 43.4 | 57.4 | 40.1 | 83.1 | 57.4 | |
| SAM 3 Detector + T-by-D as tracker2025.11 | 23.7 | 68 | 34.7 | 57.6 | 44.5 | 75.3 | 53.4 | |
| LLMDet + SAM 3 TrackerZero-shot=true2025.11 | 1.6 | 11 | 14.1 | 30.5 | 13 | 72.1 | 29.3 | |
| GLEEPrompting strategy=one NP at a time, Zero-shot=true2025.11 | 0.2 | 6 | 2.5 | 12.4 | 3.9 | 41.7 | 15.7 | |
| GLEEPrompting strategy=all NPs at once, Zero-shot=true2025.11 | 0.1 | 17 | 0.4 | 7.7 | 2.6 | 28.5 | 6.5 |