3D Visual Grounding on ScanRefer Overall
61.6Acc @ 0.5TAB
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TABLLM/VLM=Qwen3-VL-32B, w/o PC=✗2026.04 | 61.6 | 71.6 | |
| 3D-R1Supervision=Supervised, VLM=None2025.12 | 59.2 | 65.8 | |
| LLaVA-3DLLM/VLM=LLaVA-Video-7B, w/o PC=✓, Training Stages=Two-stage2026.04 | 58.6 | 63.9 | |
| GPT4SceneSupervision=Supervised, VLM=None2025.12 | 57 | 62.6 | |
| GPT4SceneLLM/VLM=Qwen2-VL-7B, w/o PC=✗2026.04 | 57 | 62.6 | |
| 3D-RSLLM/VLM=LLaVA-Next-Video, w/o PC=✗2026.04 | 56.1 | 62.9 | |
| CAPrunerBackbone=Llama-3-8B2026.04 | 56 | 61.7 | |
| 3DGraphLLMSize=8B2026.04 | 54.6 | 60.2 | |
| OpenGroundSupervision=Zero-Shot, VLM=GLM-4.5V2025.12 | 53.1 | 61.8 | |
| QuatRoPE2026.04 | 52.5 | 58.2 | |
| Video-3D-LLMLLM/VLM=LLaVA-Video-7B, w/o PC=✗2026.04 | 51.7 | 58.1 | |
| PQ3D2026.04 | 51.2 | — | |
| VG-LLMLLM/VLM=Qwen2.5-VL-7B, w/o PC=✗2026.04 | 50.9 | 57.6 | |
| Chat-SceneLLM/VLM=Vicuna-7B, w/o PC=✗2026.04 | 50.2 | 55.5 | |
| Chat-Scene-7BSize=7B2026.04 | 50.2 | 55.5 | |
| SeqVLMSupervision=Zero-Shot, VLM=Doubao-1.5-pro2025.12 | 49.6 | 55.6 | |
| SeqVLMLLM/VLM=Doubao-1.5-pro, w/o PC=✗2026.04 | 49.6 | 55.6 | |
| CAPrunerBackbone=Llama-3.2-1B2026.04 | 49.6 | 55 | |
| SPAZERSupervision=Zero-Shot, VLM=GPT-4o2025.12 | 48.8 | 57.2 | |
| SPAZERLLM/VLM=GPT-4o, w/o PC=✗2026.04 | 48.8 | 57.2 | |
| 3DGraphLLMSize=1B2026.04 | 47.5 | 52.5 | |
| TSP3DSupervision=Supervised, VLM=None2025.12 | 46.7 | 56.4 | |
| TSP3D2026.04 | 46.7 | 56.5 | |
| TABLLM/VLM=Qwen3-VL-32B, w/o PC=✓2026.04 | 46.4 | 71.2 | |
| 3D-VisTA2026.04 | 45.8 | 50.6 | |
| MA2TransVG2026.04 | 45.7 | 57.9 | |
| G3-LQLLM/VLM=-, w/o PC=✗2026.04 | 44.7 | 56 | |
| AugRefer2026.04 | 44 | 55.7 | |
| SPAR-mixLLM/VLM=InternVL-2.5-8B, w/o PC=✗2026.04 | 43.1 | 48.8 | |
| LLaVA-3DLLM/VLM=LLaVA-Video-7B, w/o PC=✓, Training Stages=One-stage2026.04 | 42.7 | 50.1 | |
| EDALLM/VLM=-, w/o PC=✗2026.04 | 42.3 | 54.6 | |
| BUTD-DETRLLM/VLM=-, w/o PC=✗2026.04 | 39.8 | 52.2 | |
| CSVGLLM/VLM=Mistral-Large-2407, w/o PC=✗2026.04 | 39.8 | 49.6 | |
| CSVG2026.04 | 39.8 | 49.6 | |
| SeeGroundSupervision=Zero-Shot, VLM=Qwen2-VL-72b2025.12 | 39.4 | 44.1 | |
| SeeGroundLLM/VLM=Qwen2-VL-72B, w/o PC=✗2026.04 | 39.4 | 44.1 | |
| SeeGround2026.04 | 39.4 | 44.1 | |
| ViewSRDSupervision=Supervised, VLM=None2025.12 | 36 | 45.4 | |
| 3DVG-TLLM/VLM=-, w/o PC=✗2026.04 | 34.5 | 45.9 | |
| FFL-3DOG2026.04 | 34 | 41.3 | |
| VLM-GrounderSupervision=Zero-Shot, VLM=GPT-4o, notes=results on selected 250 samples2025.12 | 33.5 | 48.3 | |
| Multi-View Transformerview num=42022.04 | 33.26 | 40.8 | |
| InstanceRefer2022.04 | 32.93 | 40.23 | |
| VLM-GrounderLLM/VLM=GPT-4o, w/o PC=✓2026.04 | 32.8 | 51.6 | |
| ZSVG3DSupervision=Zero-Shot, VLM=GPT-4 turbo2025.12 | 32.7 | 36.4 | |
| ZSVG3DLLM/VLM=GPT-4 turbo, w/o PC=✗2026.04 | 32.7 | 36.4 | |
| Multi-View Transformerview num=12022.04 | 31.12 | 38.33 | |
| TGNN + BERT2022.04 | 29.7 | 37.37 | |
| TGNN2022.04 | 27.92 | 34.29 | |
| ScanReferSupervision=Supervised, VLM=None2025.12 | 26.1 | 39 | |
| ScanReferLLM/VLM=-, w/o PC=✗2026.04 | 26.1 | 39 | |
| ScanRefer2022.04 | 24.32 | 37.3 | |
| VG-LLMLLM/VLM=Qwen2.5-VL-7B, w/o PC=✓2026.04 | 14.9 | 41.6 | |
| SPAR-mixLLM/VLM=InternVL-2.5-8B, w/o PC=✓2026.04 | 12.4 | 31.9 | |
| LLM-GrounderLLM/VLM=GPT-4 turbo, w/o PC=✗2026.04 | 5.3 | 17.1 |