Visually-Grounded Active View Selection on AVS-ProcTHOR (val)
93.02Existence ScoreNo Action (Target view)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| No Action (Target view)Action Model=No Action, Evaluation View=Target view2025.12 | 93.02 | 69.14 | 92.58 | 84.91 | |
| AVS Framework (SFT+RL)Backbone=Qwen2.5-VL-7B, Training Strategy=SFT+RL2025.12 | 91.47 | 69.52 | 90.17 | 83.72 | |
| AVS Framework (SFT)Backbone=Qwen2.5-VL-7B, Training Strategy=SFT2025.12 | 91.28 | 57.06 | 83.84 | 77.39 | |
| AVS Framework (RL)Backbone=Qwen2.5-VL-7B, Training Strategy=RL2025.12 | 86.82 | 65.24 | 83.41 | 78.49 | |
| Gemini-2.5-ProAction Model=Proprietary Models2025.12 | 82.95 | 52.79 | 81 | 72.25 | |
| GPT-5Action Model=Proprietary Models2025.12 | 81.01 | 55.58 | 79.69 | 72.09 | |
| Qwen2.5-VL-7BAction Model=Backbone Model, Backbone=Qwen2.5-VL-7B2025.12 | 64.34 | 29.74 | 56.55 | 50.21 | |
| Fine-EQAAction Model=EQA Framework2025.12 | 63.57 | 31.97 | 64.41 | 53.32 | |
| ViLaSRAction Model=Spatial VLMs2025.12 | 57.95 | 25.46 | 52.84 | 45.42 | |
| SpatialReasonerAction Model=Spatial VLMs2025.12 | 54.65 | 22.68 | 52.62 | 43.32 | |
| No Action (Query view)Action Model=No Action, Evaluation View=Query view2025.12 | 49.22 | 16.36 | 61.57 | 42.38 |