Object Detection on ScanNet 2D (val)
50.4AP@0.5Mask3D
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Mask3DPre-training=Masked 3D prior learning on ImageNet + ScanNet, Backbone=ViT2023.02 | 50.4 | 35.3 | 32.7 | |
| MAE-unsupIN→SNPre-training=MAE (ImageNet unsupervised) to ScanNet, Backbone=ViT2023.02 | 46.1 | 32.7 | 30.5 | |
| Pri3DPre-training=3D-based pre-training on ImageNet + ScanNet2023.02 | 43.7 | 27 | 26.3 | |
| MoCoV2-supIN→SNPre-training=MoCoV2 (ImageNet supervised) to ScanNet2023.02 | 43.5 | 26.8 | 25.8 | |
| ImageNet Pretrain (supIN)Pre-training=Supervised ImageNet2023.02 | 41.7 | 25.9 | 25.1 | |
| ScratchPre-training=None2023.02 | 32.7 | 17.7 | 16.9 |