Grounded Reasoning on VRT-Bench 1.0 (test)
66.6Comp R-LQR-Sa2VA-Qwen3VL-4B
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| R-Sa2VA-Qwen3VL-4BModel category=reasoning models, Evaluation protocol=seg evaluation2025.12 | 66.6 | 87.5 | 61.7 | 68.9 | 87.9 | 60.9 | 65.5 | 86.2 | 55.3 | 64 | 87.6 | 63 | 66.3 | 87.3 | 59.5 | |
| R-Sa2VA-InternVL-4BModel category=reasoning models, Evaluation protocol=seg evaluation2025.12 | 60.5 | 87.9 | 51.6 | 62.6 | 88.1 | 54.6 | 60.2 | 86.8 | 49.6 | 60.6 | 87.8 | 56.5 | 61.2 | 87.5 | 54 | |
| Gemini-2.5 Pro + SAM2Model category=seg models, Evaluation protocol=seg evaluation2025.12 | 36.4 | 87.7 | 40.8 | 31.5 | 89 | 50.7 | 32.5 | 86.8 | 49.7 | 34.4 | 88.6 | 53.5 | 33 | 87.5 | 49.1 | |
| Gemini-2.5 ProModel category=grounding models, Evaluation protocol=box evaluation2025.12 | 29.2 | 72 | 38.2 | 25.7 | 73.6 | 39.1 | 26.2 | 71.7 | 38.4 | 27.2 | 73.1 | 40.7 | 26.6 | 72.1 | 39 | |
| Qwen3-VL-8B-InstructModel category=grounding models, Evaluation protocol=box evaluation2025.12 | 13.6 | 88.9 | 47.3 | 14.7 | 88.3 | 56.3 | 16.5 | 86.7 | 49.2 | 14.1 | 87 | 44.8 | 14.5 | 87.5 | 49.7 | |
| Qwen3-VL-8B-Instruct + SAM2Model category=seg models, Evaluation protocol=seg evaluation2025.12 | 13.6 | 93.8 | 48.9 | 15.2 | 91.2 | 58.5 | 16.3 | 91.2 | 51.2 | 14.3 | 91.7 | 45.9 | 14.7 | 91.7 | 51.7 | |
| Qwen3-VL-4B-InstructModel category=grounding models, Evaluation protocol=box evaluation2025.12 | 6.9 | 85.5 | 53.5 | 9.3 | 88.2 | 51.2 | 12.5 | 85.4 | 52.5 | 12.4 | 85 | 54.2 | 10.2 | 86.2 | 50.6 | |
| Qwen3-VL-4B-Instruct + SAM2Model category=seg models, Evaluation protocol=seg evaluation2025.12 | 6.6 | 95.5 | 57.3 | 9.3 | 92 | 53.9 | 13 | 89.8 | 55.9 | 12.4 | 90.3 | 58.2 | 10.3 | 91.1 | 53.9 | |
| Sa2VA-Qwen3-4BModel category=seg models, Evaluation protocol=seg evaluation2025.12 | 0 | 0 | 36.7 | 0 | 0 | 40.6 | 0 | 0 | 39.1 | 0 | 0 | 39.7 | 0 | 0 | 39.4 |