Multiview Scene Graph Generation on ARKitScenes (test)
98.61Recall@1AoMSG-B-4
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AoMSG-B-4Backbone=DINOv2, Transformer decoder layers=42026.04 | 98.61 | 17.87 | 44.46 | 18.15 | |
| AoMSG-2Backbone=DINOv2, Transformer decoder layers=22026.04 | 98.5 | 21.02 | 49.18 | 21.39 | |
| AoMSG-1Backbone=DINOv2, Transformer decoder layers=12026.04 | 98.47 | 24.87 | 55.5 | 25.37 | |
| HSGBackbone=DINOv22026.04 | 98.39 | 33.17 | 45.52 | 33.51 | |
| AoMSG-4 (baseline)Backbone=DINOv2, Transformer decoder layers=4, mode=baseline2026.04 | 98.33 | 21.02 | 42.79 | 21.32 | |
| SepMSG-DirectBackbone=DINOv2, Projection=direct2026.04 | 97.71 | 33.19 | 48.58 | 33.67 | |
| SepMSG-MLPBackbone=DINOv2, Projection=MLP2026.04 | 97.39 | 20.51 | 58.63 | 21.01 | |
| SepMSG-LinearBackbone=DINOv2, Projection=linear2026.04 | 97.3 | 20.02 | 55.67 | 20.48 |