Active Question Answering on OpenEQA v1 (HM3D)
85.1LLM-MatchHuman baseline
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Human baseline2025.12 | 85.1 | — | |
| R42025.12 | 74 | 63.37 | |
| 3D-Mem2025.12 | 52.6 | 42 | |
| GPT-4V2025.12 | 41.8 | 7.5 | |
| GPT-4 w/ LLaVA-1.5Vision-Language Model=LLaVA-1.52025.12 | 38.1 | 7 | |
| GPT-42025.12 | 35.5 | — | |
| GPT-4 w/ CGRepresentation=Concept Graphs2025.12 | 34.4 | 6.5 | |
| GPT-4 w/ SVMRepresentation=Sparse Voxel Map2025.12 | 34.2 | 6.4 | |
| LLaMA-2 w/ LLaVA-1.5Vision-Language Model=LLaVA-1.52025.12 | 30.9 | 5.9 | |
| LLaMA-2 w/ SVMRepresentation=Sparse Voxel Map2025.12 | 29.9 | 5.5 | |
| LLaMA-22025.12 | 29 | — | |
| LLaMA-2 w/ CGRepresentation=Concept Graphs2025.12 | 23.9 | 4.3 |