Visual Grounding on Multi3DRefer (F1@0.25 and F1@0.5)
63.8F1@0.25CAIRN
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CAIRNLLM=Qwen3-8B2026.07 | 63.8 | 58.8 | |
| 3DGraphLLMLLM=Qwen3-8B2026.07 | 62.9 | 58.1 | |
| Inst3D-LMMLLM=Qwen3-8B2026.07 | 62.4 | 57.9 | |
| Chat-SceneLLM=Qwen3-8B2026.07 | 61.3 | 56.7 | |
| Inst3D-LMMLLM=Vicuna-7B-v1.52026.07 | 58.3 | 53.5 | |
| Chat-SceneLLM=Vicuna-7B-v1.52026.07 | 57.1 | 52.4 | |
| GPT4Scene-HDLLM=Qwen2-VL-7B2026.07 | 53.7 | 50 | |
| PQ3DLLM=None2026.07 | 50.1 | — | |
| Grounded 3D-LLMLLM=Tiny-Vicuna-1B2026.07 | 45.2 | 40.6 | |
| M3DRef-CLIPLLM=None2026.07 | 42.8 | 38.4 |