Hierarchical Scene Parsing on SceneParser-Bench object-level
78.3L1 PrecisionRex-Omni Stitching
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Rex-Omni StitchingCategory=Rex-Omni based methods2026.05 | 78.3 | 24.2 | 37 | 76.1 | 13.1 | 22.4 | 0 | 0 | 0 | 15 | |
| SceneParser2026.05 | 76.8 | 70.4 | 73.5 | 43.8 | 42.1 | 42.9 | 30 | 28.7 | 29.3 | 66.3 | |
| Gemini 3.1 ProAccess=Closed-source MLLMs2026.05 | 67.3 | 48.4 | 56.3 | 11 | 6.4 | 8.1 | 0.2 | 0.1 | 0.1 | 28.6 | |
| Qwen2.5-VL-72BAccess=Open-source MLLMs2026.05 | 55.2 | 34.1 | 42.1 | 4.9 | 2.4 | 3.2 | 0.1 | 0 | 0.1 | 29.5 | |
| SEED2.0Access=Open-source MLLMs2026.05 | 41.6 | 34.8 | 37.9 | 5.7 | 4.4 | 5 | 0 | 0 | 0 | 26 | |
| Qwen3.5-35BAccess=Open-source MLLMs2026.05 | 32.9 | 48.8 | 39.3 | 3.8 | 9.6 | 5.4 | 0 | 0.1 | 0 | 60.8 | |
| Qwen3.5-plusAccess=Closed-source MLLMs2026.05 | 25.9 | 52.2 | 34.6 | 3.1 | 10 | 4.7 | 0 | 0.1 | 0.1 | 64.4 | |
| GPT-5.5Access=Closed-source MLLMs2026.05 | 23.8 | 19.5 | 21.5 | 1.3 | 3 | 1.8 | 0 | 0 | 0 | 25.3 | |
| Qwen3-VL-235BAccess=Open-source MLLMs2026.05 | 21.2 | 41.8 | 28.1 | 3.1 | 8.7 | 4.6 | 0 | 0 | 0 | 49.6 | |
| Qwen3.5-4BAccess=Open-source MLLMs2026.05 | 2.5 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |