Object Detection on ScanNet (val)
52.2mAP@0.5HUNet
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HUNetBackbone=3DETR, Protocol=Fine-Tuning2025.04 | 52.2 | 71.3 | |
| Point-M2AEBackbone=3DETR, Protocol=Fine-Tuning2025.04 | 48.3 | 66.3 | |
| Dataset-specific (Oracle)Training Data=ScanNet training set, Backbone=ResNet-502021.02 | 44.7 | — | |
| Point-MAEBackbone=3DETR, Protocol=Fine-Tuning2025.04 | 40.6 | 63.4 | |
| MaskPointBackbone=3DETR, Protocol=Fine-Tuning2025.04 | 40.6 | 63.4 | |
| HUNetBackbone=3DETR, Protocol=Linear Probing2025.04 | 40.2 | 65.6 | |
| Point-BertBackbone=3DETR, Protocol=Fine-Tuning2025.04 | 38.3 | 61 | |
| 3DETRBackbone=3DETR2025.04 | 37.9 | 62.1 | |
| Partitioned DetectorStrategy=Partitioned, Backbone=ResNet-502021.02 | 32.2 | — | |
| EnsembleStrategy=Ensemble, Backbone=ResNet-502021.02 | 30.1 | — | |
| Unified detector (UniDet)Strategy=Unified (retrained), Backbone=ResNet-502021.02 | 29.8 | — | |
| Objects365 modelTraining Data=Objects365, Backbone=ResNet-502021.02 | 24.9 | — | |
| OpenImages modelTraining Data=OpenImages, Backbone=ResNet-502021.02 | 24.2 | — | |
| COCO modelTraining Data=COCO, Backbone=ResNet-502021.02 | 17.4 | — | |
| Mapillary modelTraining Data=Mapillary, Backbone=ResNet-502021.02 | 0 | — |