3D Object Detection on ScanNet V2 (val)
78.1mAP@0.25ISBNet
Evaluation results
| Method | Links | ||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ISBNet2023.03 | 78.1 | 62 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MV-DETR2024.08 | 78 | 65.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| V-DETRProtocol=Single Run, TTA=true2023.08 | 77.8 | 66 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| V-DETR2024.08 | 77.8 | 65.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| V-DETRProtocol=Single Run, TTA=false2023.08 | 77.4 | 65 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| V-DETRProtocol=Average 25x trials, TTA=true2023.08 | 77 | 65.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OneFormer3D2023.11 | 76.9 | 65.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OneFormer3D2024.08 | 76.9 | 65.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| V-DETRProtocol=Average 25x trials, TTA=false2023.08 | 76.8 | 64.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Swin3D-L+CAGroup3DBackbone=Swin3D-L2024.08 | 76.4 | 63.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IAELearning Paradigm=Self-supervised, Backbone=CAGroup3D2022.01 | 76.1 | 62 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IAEBase Architecture=CAGroup3D, Pre-training Setting=Self-supervised2025.12 | 76.1 | 62 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAGroup3Dmulti-sensor=false2022.10 | 75.1 | 61.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAGroup3DProtocol=Single Run2023.08 | 75.1 | 61.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAGroup3DLearning Paradigm=Supervised2022.01 | 75.1 | 61.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAGroup3DDesign Task=det2023.09 | 75.1 | 61.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAGroup3DPresented at=NeurIPS'222023.11 | 75.1 | 61.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAGroup3D2024.08 | 75.1 | 61.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAGroup3DPre-training Setting=Supervised2025.12 | 75.1 | 61.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAGroup3DProtocol=Average 25x trials2023.08 | 74.5 | 60.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SPGroup3DAggregation=Best2023.12 | 74.3 | 59.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SPGroup3D2024.08 | 74.3 | 59.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Mask-Attention-Free TransformerDesign Task=inst2023.09 | 73.5 | 63.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SPGroup3DAggregation=Average (25 trials)2023.12 | 73.5 | 58.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAGroup3D*Presented at=NeurIPS'22, Note=One-stage result from ablation study2023.12 | 73.2 | 57.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Point-GCCTraining Protocol=Fine-tuning with pre-training, Backbone=TR3D2023.05 | 73.1 | 59.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Point-GCC+TR3DAugmentation=Point-GCC2024.08 | 73.1 | 59.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TR3DTraining Protocol=Supervised Only2023.05 | 72.9 | 59.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TR3DTraining Protocol=Self-supervised Pre-training (Baseline)2023.05 | 72.9 | 59.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TR3DProtocol=Single Run2023.08 | 72.9 | 59.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TR3DPresented at=ICIP'232023.11 | 72.9 | 59.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TR3D2024.08 | 72.9 | 59.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ORACLEDescription=Upper bound (full annotation), Base detector=CAGroup3D2025.03 | 72.65 | 58.81 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Ours (GroupFree3D)Pre-trained=Yes, Backbone=GroupFree3D2024.10 | 72.3 | 55.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TR3DProtocol=Average 25x trials2023.08 | 72 | 57.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TR3DPresent at=ICIP, Paradigm=Fully Supervised2025.03 | 72 | 57.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Uni3DETRModality=Point Clouds2023.10 | 71.7 | 58.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SoftGroup2023.03 | 71.6 | 59.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SoftGroupPresented at=CVPR'222023.11 | 71.6 | 59.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DPresented at=ECCV'22, multi-sensor=false2022.10 | 71.5 | 57.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DTraining Protocol=Supervised Only2023.05 | 71.5 | 57.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DProtocol=Single Run2023.08 | 71.5 | 57.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DLearning Paradigm=Supervised2022.01 | 71.5 | 57.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DDesign Task=det2023.09 | 71.5 | 57.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DModality=Point Clouds2023.10 | 71.5 | 57.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DPresented at=ECCV'222023.11 | 71.5 | 57.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DPresented at=ECCV'22, Aggregation=Best2023.12 | 71.5 | 57.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DPre-training Setting=Supervised2025.12 | 71.5 | 57.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BrTBackbone=PointNet++w2x, Image Backbone=ViT-B/16, RGB=true2022.10 | 71.3 | 55.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFree (12, 512) with RepSurf-U²Backbone=RepSurf-U², #Params=23.61M, Infer Speed=195ms2022.05 | 71.2 | 54.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HyperDet3DPresented at=CVPR 22, multi-sensor=false2022.10 | 70.9 | 57.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HyperDet3DProtocol=Single Run2023.08 | 70.9 | 57.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HyperDet3DDesign Task=det2023.09 | 70.9 | 57.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HyperDet3DModality=Point Clouds2023.10 | 70.9 | 57.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HyperDet3DPresented at=CVPR 222023.12 | 70.9 | 57.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TokenFusionPresented at=CVPR 22, multi-sensor=true2022.10 | 70.8 | 54.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TokenFusionModality=Point Clouds + RGB2023.10 | 70.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DProtocol=Average 25x trials2023.08 | 70.7 | 56 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DPresented at=ECCV'22, Aggregation=Average (25 trials)2023.12 | 70.7 | 56 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FCAF3DPresent at=ECCV, Paradigm=Fully Supervised2025.03 | 70.7 | 56 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetBackbone=PointNet++w2x, Input modality=Geometric, Number of rays=66, Number of object candidates=512, Result reporting=Best2022.04 | 70.6 | 55.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetPresented at=CVPR 22, multi-sensor=false2022.10 | 70.6 | 55.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetProtocol=Single Run2023.08 | 70.6 | 55.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetDesign Task=det2023.09 | 70.6 | 55.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetModality=Point Clouds2023.10 | 70.6 | 55.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetPresented at=CVPR 22, Aggregation=Best2023.12 | 70.6 | 55.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFree (12, 512) with RepSurf-T²Backbone=RepSurf-T², #Params=23.60M, Infer Speed=194ms2022.05 | 70.4 | 54.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetBackbone=PointNet++, Input modality=Geometric, Number of rays=66, Number of object candidates=256, Result reporting=Best2022.04 | 70.2 | 54.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Mask3DDesign Task=inst2023.09 | 70.2 | 56.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetBackbone=PointNet++w2x, Input modality=Geometric, Number of rays=66, Number of object candidates=512, Result reporting=Average2022.04 | 69.9 | 54.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetProtocol=Average 25x trials2023.08 | 69.9 | 54.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BrTBackbone=PointNet++, Image Backbone=ViT-S/16, RGB=true2022.10 | 69.7 | 53 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetBackbone=PointNet++, Input modality=Geometric, Number of rays=66, Number of object candidates=256, Result reporting=Average2022.04 | 69.6 | 53.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RBGNetPresented at=CVPR 22, Aggregation=Average (25 trials)2023.12 | 69.6 | 54.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFree3D*+DisARMImplementation Platform=MMDetection3D, Layers=L12, Proposals=O512, Width=w2x, Method Enhancement=DisARM2022.03 | 69.3 | 53.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PBNetPresented at=ICCV'232023.11 | 69.3 | 60.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-Free (24-L) + YOLOSBackbone=PointNet++w2x, RGB=true2022.10 | 69.2 | 52.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-Free-3Dbackbone=PointNet++w2x, decoder depth (L)=12, number of object candidates (O)=5122021.04 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-Free (L12, O512)backbone=PointNet++w2x, layers=12, object candidates=5122021.04 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 5.17 | — | — | — | — | — | |
| Group-freeBackbone=PointNet++w2x, Input modality=Geometric, Result reporting=Best2022.04 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFree (12, 512)Backbone=PointNet++², #Params=23.60M, Infer Speed=193ms2022.05 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-FreeBackbone=PointNet++w2x, Layers=24-L, RGB=false2022.10 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-freePresented at=ICCV'21, multi-sensor=false2022.10 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-FreeProtocol=Single Run2023.08 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFreeLearning Paradigm=Supervised2022.01 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-freeDesign Task=det2023.09 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFreeModality=Point Clouds2023.10 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFreePresented at=ICCV'21, Aggregation=Best2023.12 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Bridge3D (GroupFree3D)Pre-trained=Yes, Backbone=GroupFree3D2024.10 | 69.1 | 51.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFreePre-training Setting=Supervised2025.12 | 69.1 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFree3D + HPE_MLPMMDetection3D implementation=true2026.03 | 69.1 | 53 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFree 6,256BackBone=X-3D, #Param=12.30, GFLOPS=7.12024.04 | 69 | 51.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFree3D + X-3D2026.03 | 69 | 51.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-Free-3Dbackbone=PointNet++w2x, decoder depth (L)=12, number of object candidates (O)=2562021.04 | 68.8 | 52.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-Free (L12, O256)backbone=PointNet++w2x, layers=12, object candidates=2562021.04 | 68.8 | 52.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 5.23 | — | — | — | — | — | |
| GroupFree (6, 256) with RepSurf-UBackbone=RepSurf-U, #Params=11.50M, Infer Speed=150ms2022.05 | 68.8 | 50.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupFree 6,256BackBone=RepSurf-U, #Param=11.50, GFLOPS=6.82024.04 | 68.8 | 50.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-Free (24-L) + Faster R-CNNBackbone=PointNet++w2x, RGB=true2022.10 | 68.7 | 52.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-freeBackbone=PointNet++w2x, Input modality=Geometric, Result reporting=Average2022.04 | 68.6 | 51.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Group-FreeProtocol=Average 25x trials2023.08 | 68.6 | 51.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |