Visually-Grounded Active View Selection on AVS-HM3D (val)
86.67ExistenceNo Action (Target view)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| No Action (Target view)Action Model=No Action, Evaluation View=Target view2025.12 | 86.67 | 77.5 | 80 | 74.48 | 71.79 | 78.09 | |
| AVS Framework (RL)Backbone=Qwen2.5-VL-7B, Training Strategy=RL2025.12 | 81.25 | 70 | 72.97 | 69.17 | 60 | 70.68 | |
| GPT-5Action Model=Proprietary Models2025.12 | 76.67 | 54.62 | 65.95 | 65.83 | 60 | 64.91 | |
| AVS Framework (SFT+RL)Backbone=Qwen2.5-VL-7B, Training Strategy=SFT+RL2025.12 | 74.58 | 71.54 | 73.78 | 70.83 | 62.78 | 70.7 | |
| Gemini-2.5-ProAction Model=Proprietary Models2025.12 | 74.17 | 59.23 | 60.81 | 64.17 | 60 | 63.67 | |
| SpatialReasonerAction Model=Spatial VLMs2025.12 | 70.42 | 50 | 47.3 | 37.5 | 40 | 49.04 | |
| Fine-EQAAction Model=EQA Framework2025.12 | 70 | 52.31 | 52.7 | 54.17 | 44.44 | 54.72 | |
| ViLaSRAction Model=Spatial VLMs2025.12 | 68.33 | 48.46 | 52.7 | 50 | 50 | 53.9 | |
| No Action (Query view)Action Model=No Action, Evaluation View=Query view2025.12 | 67.5 | 56.15 | 54.59 | 66.67 | 48.33 | 58.65 | |
| AVS Framework (SFT)Backbone=Qwen2.5-VL-7B, Training Strategy=SFT2025.12 | 67.5 | 70.77 | 62.16 | 66.67 | 55.56 | 64.53 | |
| Qwen2.5-VL-7BAction Model=Backbone Model, Backbone=Qwen2.5-VL-7B2025.12 | 66.25 | 46.15 | 66.49 | 49.17 | 46.86 | 54.98 |