3D Object Detection on SUN RGB-D (mAP@0.25/0.5)
67.9mAP@0.25GroupFree3D + Bridge3D
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GroupFree3D + Bridge3DPre-trained=true2023.05 | 67.9 | 48.5 | |
| IAEBackbone=CAGroup3D2025.12 | 67.6 | 51.3 | |
| MVNetDetection Architecture=CAGroup3D2025.12 | 67.6 | 51.7 | |
| MaskFeat3DBackbone=CAGroup3D2025.12 | 67.2 | 51 | |
| ConDaFormerbaseline=CAGroup3D2023.12 | 67.1 | 49.9 | |
| CAGroup3D2023.12 | 66.8 | 50.2 | |
| CAGroup3D2025.12 | 66.8 | 50.2 | |
| GroupFree3D + PIMAEPre-trained=true2023.05 | 65 | 46.8 | |
| OneDet3DTraining setup=multi-dataset training2024.11 | 65 | 51.3 | |
| ConDaFormerbaseline=FCAF3D2023.12 | 64.9 | 48.8 | |
| PillarDETR2026.06 | 64.8 | 46.5 | |
| GroupFree 6,256BackBone=X-3D, #Param=12.30, GFLOPS=7.12024.04 | 64.5 | 46.9 | |
| GroupFree 6,256BackBone=RepSurf-U, #Param=11.50, GFLOPS=6.82024.04 | 64.3 | 45.9 | |
| FCAF3D2021.12 | 64.2 | 48.9 | |
| FCAF3D2023.12 | 64.2 | 48.9 | |
| FCAF3D2025.12 | 64.2 | 48.9 | |
| GroupFree3D + Point-MAEPre-trained=true2023.05 | 63.9 | 46.1 | |
| FCAF3DTraining setup=single-dataset training2024.11 | 63.8 | 48.2 | |
| ImVoteNetBackBone=PointNet++2024.04 | 63.4 | — | |
| OneDet3DTraining setup=single-dataset training2024.11 | 63.2 | 48.7 | |
| GroupFreePresented at=ICCV'212021.12 | 63 | 45.2 | |
| GroupFree3DPre-trained=false2023.05 | 63 | 45.2 | |
| GroupFree2023.12 | 63 | 45.2 | |
| GroupFree 6,256BackBone=PointNet++, #Param=11.49, GFLOPS=6.72024.04 | 63 | 45.2 | |
| GroupFree2025.12 | 63 | 45.2 | |
| GroupFreeTraining setup=single-dataset training2024.11 | 63 | 45.2 | |
| GroupFree3D2026.06 | 63 | 45.2 | |
| VENetPresented at=ICCV'212021.12 | 62.5 | 39.2 | |
| VENet2023.12 | 62.5 | 39.2 | |
| GroupFree3D + Bridge3D (from scratch)Pre-trained=false2023.05 | 62.2 | 45 | |
| MVNetDetection Architecture=VoteNet2025.12 | 62 | 39.6 | |
| 3DETR + Bridge3DPre-trained=true2023.05 | 61.8 | 35.9 | |
| RDFC-GANBase Detector=H3DNet2023.06 | 61.75 | 40.63 | |
| HGNetPresented at=CVPR 202021.12 | 61.6 | — | |
| DepthContrastPre-trained=true2023.05 | 61.6 | 35.5 | |
| HGNet2023.12 | 61.6 | — | |
| HGNetArch.=VoteNet + GraphConv2021.09 | 61.6 | 34.4 | |
| BRNetPresented at=CVPR'212021.12 | 61.1 | 43.7 | |
| BRNet2023.12 | 61.1 | 43.7 | |
| BRNetBackBone=PointNet++2024.04 | 61.1 | 43.7 | |
| BRNet2025.12 | 61.1 | 43.7 | |
| RDF-GANBase Detector=H3DNet2023.06 | 61.03 | 39.71 | |
| RDFC-GANBase Detector=VoteNet2023.06 | 61.02 | 37.47 | |
| MaskFeat3DBackbone=VoteNet2025.12 | 61 | 36.5 | |
| RDF-GANBase Detector=VoteNet2023.06 | 60.64 | 37.28 | |
| MM-3DScene + VoteNetPre-training Strategy=MM-3DScene, Backbone/Detector=VoteNet2022.12 | 60.6 | 37.3 | |
| MM-3DSceneDetection Architecture=VoteNet2025.12 | 60.6 | 37.3 | |
| GTBase Detector=H3DNet2023.06 | 60.51 | 39.22 | |
| DepthContrast2025.12 | 60.4 | 35.4 | |
| IAEBackbone=VoteNet2025.12 | 60.4 | 36 | |
| DepthContrastDetection Architecture=VoteNet2025.12 | 60.4 | 35.4 | |
| DepthContrastBackbone=VoteNet2025.12 | 60.4 | 35.4 | |
| DeepLidarBase Detector=H3DNet2023.06 | 60.35 | 39.16 | |
| DPCoPre-trained=true2023.05 | 60.2 | 35.5 | |
| H3DNetBase Detector=H3DNet2023.06 | 60.11 | 39.04 | |
| H3DNetPresented at=ECCV'202021.12 | 60.1 | 39 | |
| H3DNet2023.12 | 60.1 | 39 | |
| H3DNetBackBone=4x PointNet++2024.04 | 60.1 | 39 | |
| H3DNetArch.=VoteNet + 3D primitives2021.09 | 60.1 | 39 | |
| H3DNetTraining setup=single-dataset training2024.11 | 60.1 | 39 | |
| 3DETR + PIMAEPre-trained=true2023.05 | 59.9 | 33.7 | |
| MLCVNetPresented at=CVPR 202021.12 | 59.8 | — | |
| MLCVNet2023.12 | 59.8 | — | |
| MLCVNet2025.12 | 59.8 | — | |
| DeepLidarBase Detector=VoteNet2023.06 | 59.73 | 35.49 | |
| GTBase Detector=VoteNet2023.06 | 59.44 | 36.3 | |
| RandomRooms + VoteNetPre-training Strategy=RandomRooms, Backbone/Detector=VoteNet2022.12 | 59.2 | 35.4 | |
| RandomRooms2025.12 | 59.2 | 35.4 | |
| RandomRoomsDetection Architecture=VoteNet2025.12 | 59.2 | 35.4 | |
| RandomRoomsBackbone=VoteNet2025.12 | 59.2 | 35.4 | |
| 3DETRPresented at=ICCV'212021.12 | 59.1 | 32.7 | |
| 3DETR2023.12 | 59.1 | 32.7 | |
| VoteNetBackBone=PointNet++2024.04 | 59.1 | 35.8 | |
| 3DETRBackBone=Transformer2024.04 | 59.1 | 32.7 | |
| 3DETR2025.12 | 59.1 | 32.7 | |
| VoteNetBase Detector=VoteNet2023.06 | 59.07 | 35.77 | |
| 3DETR-mArch.=Tx.2021.09 | 59 | 32.7 | |
| VoteNetArch.=VoteNet2021.09 | 58.3 | 33.4 | |
| STRL2025.12 | 58.2 | 35 | |
| STRLDetection Architecture=VoteNet2025.12 | 58.2 | 35 | |
| STRLBackbone=VoteNet2025.12 | 58.2 | 35 | |
| 3DETRPre-trained=false2023.05 | 58 | 30.3 | |
| VoteNetPresented at=ICCV'192021.12 | 57.7 | — | |
| VoteNetPre-training Strategy=from scratch, Backbone/Detector=VoteNet2022.12 | 57.7 | 32.9 | |
| VoteNetPre-trained=false2023.05 | 57.7 | 32.9 | |
| VoteNet2023.12 | 57.7 | — | |
| VoteNet2025.12 | 57.7 | 32.9 | |
| VoteNetTraining setup=single-dataset training2024.11 | 57.7 | 35.7 | |
| VoteNet2026.06 | 57.7 | 32.9 | |
| 3DETR + Bridge3D (from scratch)Pre-trained=false2023.05 | 57.6 | 31.9 | |
| PointContrast + VoteNetPre-training Strategy=PointContrast, Backbone/Detector=VoteNet2022.12 | 57.5 | 34.8 | |
| PointContrastPre-trained=true2023.05 | 57.5 | 34.5 | |
| PointContrast2025.12 | 57.5 | 34.8 | |
| PointContrastDetection Architecture=VoteNet2025.12 | 57.5 | 34.8 | |
| PointContrastBackbone=VoteNet2025.12 | 57.5 | 34.8 | |
| 3DETRArch.=Tx.2021.09 | 56.8 | 30.1 | |
| UVTRTraining setup=single-dataset training2024.11 | 55 | 33.2 | |
| F-PointNetBackbone=F-PointNet2022.12 | 54 | — | |
| BoxNetArch.=BoxNet2021.09 | 52.4 | 25.1 | |
| COGBackbone=COG2022.12 | 47.6 | — |