Referring Video Object Segmentation on Ambi-RVOS (Difficult)
45.5J&F ScoreIC-Seg-8B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| IC-Seg-8BType=MLLM-based methods, Parameters=8B2026.05 | 45.5 | 42.6 | 48.3 | |
| IC-Seg-4BType=MLLM-based methods, Parameters=4B2026.05 | 40.3 | 38 | 42.6 | |
| Qwen3-VL-8B*Type=MLLM-based methods, Parameters=8B, Baseline=True2026.05 | 28.2 | 25.8 | 30.6 | |
| Qwen3-VL-4B*Type=MLLM-based methods, Parameters=4B, Baseline=True2026.05 | 24.5 | 23.1 | 26 | |
| UniPixel-7BType=MLLM-based methods, Parameters=7B2026.05 | 20.3 | 16.6 | 24 | |
| GLUS-7BType=MLLM-based methods, Parameters=7B2026.05 | 18.1 | 14.7 | 21.6 | |
| Veason-R1-3BType=MLLM-based methods, Parameters=3B2026.05 | 16 | 14 | 17.9 | |
| SAM3Type=Specialist RVOS models2026.05 | 15.8 | 14.7 | 17 | |
| ReferDINOType=Specialist RVOS models2026.05 | 13.4 | 11.5 | 15.3 | |
| RGA3-7BType=MLLM-based methods, Parameters=7B2026.05 | 12.8 | 9.4 | 16.3 | |
| SAMWISEType=Specialist RVOS models2026.05 | 11.7 | 9.4 | 14 |