3D Object Classification on ScanObjectNN PB_T50_RS
91OAMVNet-L
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MVNet-LPre-training Dataset=Objaverse, Feature Dimension=7682025.12 | 91 | — | — | |
| ReCon#P (M)=43.6, #F (G)=5.3, Training Paradigm=with Pretrained Cross-Modal Teacher Representation Learning2023.07 | 90.63 | — | — | |
| I2P-MAE#P (M)=12.9, #F (G)=3.6, Training Paradigm=with Pretrained Cross-Modal Teacher Representation Learning2023.07 | 90.11 | — | — | |
| PointSD2026.06 | 90.1 | — | — | |
| PRISMcombination=with Point-MAE2026.06 | 89.9 | — | — | |
| PRISM2026.06 | 89.7 | — | — | |
| 3D-JEPAPre-training Epochs=300, Training Paradigm=SSRL2024.09 | 89.52 | — | — | |
| PointMLP + JM3Dpre-training=JM3D, voting strategy=true2023.08 | 89.5 | 88.7 | — | |
| P2P2023.08 | 89.3 | — | — | |
| VPP w/ vot.#P (M)=22.1, #F (G)=4.8, Training Paradigm=with Self-Supervised Representation Learning, Voting Strategy=Yes2023.07 | 89.28 | — | — | |
| PointMLP + JM3Dpre-training=JM3D2023.08 | 89.2 | 88.4 | — | |
| HPENetParam.=1.7, FLOPs=2.2, TP=27002026.03 | 88.9 | 87.6 | — | |
| HPENet V2-SParam.=1.5, FLOPs=0.8, TP=3989, best single-run=true2026.03 | 88.9 | 87.4 | — | |
| PointMLP + ULIPpre-training=ULIP2023.08 | 88.8 | 87.8 | — | |
| Point-DAE2026.06 | 88.7 | — | — | |
| VPP w/o vot.#P (M)=22.1, #F (G)=4.8, Training Paradigm=with Self-Supervised Representation Learning, Voting Strategy=No2023.07 | 88.65 | — | — | |
| 3D-JEPAPre-training Epochs=150, Training Paradigm=SSRL2024.09 | 88.65 | — | — | |
| MaskFeat3DBackbone=PointNeXt2025.12 | 88.6 | — | — | |
| HPENet (SIN)Param.=1.7, FLOPs=2.2, TP=24552026.03 | 88.4 | 86.9 | — | |
| HPENet V2-SParam.=1.5, FLOPs=0.8, TP=39892026.03 | 88.4 | 86.9 | — | |
| ACT#P (M)=22.1, #F (G)=4.8, Training Paradigm=with Pretrained Cross-Modal Teacher Representation Learning2023.07 | 88.21 | — | — | |
| ACTPre-training Epochs=300, Training Paradigm=SSRL2024.09 | 88.21 | — | — | |
| PointNeXt2025.12 | 88.1 | — | — | |
| PointMetaBase-SParam.=1.4, FLOPs=0.6, TP=31402026.03 | 87.9 | 86.2 | — | |
| MVNet-BPre-training Dataset=Objaverse2025.12 | 87.8 | — | — | |
| PointVectorParam.=1.6, FLOPs=1.7, TP=14842026.03 | 87.8 | 86.2 | — | |
| PointNeXt#P (M)=1.4, #F (G)=3.6, Training Paradigm=Supervised Learning Only2023.07 | 87.7 | — | — | |
| MaskFeat3DBackbone=PointViT2025.12 | 87.7 | — | — | |
| PointNeXt-SParam.=1.4, FLOPs=1.6, TP=40522026.03 | 87.7 | 85.8 | — | |
| PointDif2026.06 | 87.6 | — | — | |
| Point2VecPre-training Epochs=300, Training Paradigm=SSRL2024.09 | 87.5 | — | — | |
| MVNet-BPre-training Dataset=ShapeNet2025.12 | 86.7 | — | — | |
| Point-M2AE2026.06 | 86.5 | — | — | |
| Point-M2AE#P (M)=14.8, #F (G)=3.6, Training Paradigm=with Self-Supervised Representation Learning2023.07 | 86.43 | — | — | |
| Point-M2AEPre-training Epochs=300, Training Paradigm=SSRL2024.09 | 86.43 | — | — | |
| ULIPPre-training Dataset=ShapeNet2025.12 | 86.4 | — | — | |
| RepSurf-Uscale=2x2023.08 | 86 | — | — | |
| PointMLP2023.08 | 85.7 | 84.4 | — | |
| MaskSurfelPre-training Dataset=ShapeNet2025.12 | 85.7 | — | — | |
| PointMLPParam.=13.2, FLOPs=31.4, TP=3702026.03 | 85.7 | 84.4 | — | |
| PointMLP#P (M)=12.6, #F (G)=31.4, Training Paradigm=Supervised Learning Only2023.07 | 85.4 | — | — | |
| PointMLP2025.12 | 85.4 | — | — | |
| PointMAE2023.08 | 85.2 | — | — | |
| PointMAEPre-training Dataset=ShapeNet2025.12 | 85.2 | — | — | |
| Point-MAE2026.06 | 85.2 | — | — | |
| Point-MAE#P (M)=22.1, #F (G)=4.8, Training Paradigm=with Self-Supervised Representation Learning2023.07 | 85.18 | — | — | |
| Point-MAEPre-training Epochs=300, Training Paradigm=SSRL2024.09 | 85.18 | — | — | |
| PTv3training=from scratch2026.06 | 84.9 | — | — | |
| RepSurf-U2023.08 | 84.6 | — | — | |
| MaskDiscrPre-training Dataset=ShapeNet2025.12 | 84.3 | — | — | |
| 3D-OAEPre-training Epochs=300, Training Paradigm=SSRL2024.09 | 83.17 | — | — | |
| PointBERT2023.08 | 83.1 | — | — | |
| PointBERTPre-training Dataset=ShapeNet2025.12 | 83.1 | — | — | |
| Point-BERT2026.06 | 83.1 | — | — | |
| Point-BERT#P (M)=22.1, #F (G)=4.8, Training Paradigm=with Self-Supervised Representation Learning2023.07 | 83.07 | — | — | |
| Point-BERTPre-training Epochs=300, Training Paradigm=SSRL2024.09 | 83.07 | — | — | |
| MVTN2023.08 | 82.8 | — | — | |
| MVTNParams (M)=3.5, FLOPs (G)=1.8, Throughput (ins./sec.)=2362022.06 | 82.8 | — | — | |
| PRANetParam.=2.32026.03 | 82.1 | 79.1 | — | |
| MaskFeat3DBackbone=MinkowskiNet2025.12 | 80.8 | — | — | |
| DRNet2022.11 | 80.3 | — | — | |
| MinkowskiNet2025.12 | 80.1 | — | — | |
| Transformer#P (M)=22.1, #F (G)=4.8, Training Paradigm=with Self-Supervised Representation Learning2023.07 | 79.11 | — | — | |
| TransformerPre-training Epochs=300, Training Paradigm=SSRL2024.09 | 79.11 | — | — | |
| PointCNN2022.11 | 78.5 | — | — | |
| PointCNN#P (M)=0.6, Training Paradigm=Supervised Learning Only2023.07 | 78.5 | — | — | |
| PointCNNParams (M)=0.6, Throughput (ins./sec.)=442022.06 | 78.5 | 75.1 | — | |
| PointCNN2025.12 | 78.5 | — | — | |
| PointCNNParam.=0.62026.03 | 78.5 | 75.1 | — | |
| DGCNNFLOPs (G)=4.8, TP (ins./sec)=7352022.11 | 78.1 | — | — | |
| DGCNN#P (M)=1.8, #F (G)=2.4, Training Paradigm=Supervised Learning Only2023.07 | 78.1 | — | — | |
| DGCNN2023.08 | 78.1 | 73.6 | — | |
| DGCNNPre-training Epochs=N/A, Training Paradigm=Supervised Learning Only2024.09 | 78.1 | — | — | |
| DGCNNParams (M)=1.8, FLOPs (G)=4.8, Throughput (ins./sec.)=4022022.06 | 78.1 | 73.6 | — | |
| DGCNN2025.12 | 78.1 | — | — | |
| DGCNNParam.=1.8, FLOPs=4.8, TP=10932026.03 | 78.1 | 73.6 | — | |
| PointNet++FLOPs (G)=1.7, TP (ins./sec)=26552022.11 | 77.9 | — | — | |
| PointNet++#P (M)=1.5, #F (G)=1.7, Training Paradigm=Supervised Learning Only2023.07 | 77.9 | — | — | |
| PointNet++mode=ssg2023.08 | 77.9 | 75.4 | — | |
| PointNet++Pre-training Epochs=N/A, Training Paradigm=Supervised Learning Only2024.09 | 77.9 | — | — | |
| PointNet++Params (M)=1.5, FLOPs (G)=1.7, Throughput (ins./sec.)=18722022.06 | 77.9 | 75.4 | — | |
| PointNet++2025.12 | 77.9 | — | — | |
| PointNet++Param.=1.5, FLOPs=1.7, TP=30192026.03 | 77.9 | 75.4 | — | |
| TransformerEvaluation Protocol=from scratch2025.12 | 77.2 | — | — | |
| PointNet2023.08 | 68.2 | 63.4 | — | |
| PointNetParams (M)=3.5, FLOPs (G)=0.9, Throughput (ins./sec.)=42122022.06 | 68.2 | 63.4 | — | |
| PointNetParam.=3.5, FLOPs=0.9, TP=68462026.03 | 68.2 | 63.4 | — | |
| PointNet#P (M)=3.5, #F (G)=0.5, Training Paradigm=Supervised Learning Only2023.07 | 68 | — | — | |
| PointNetPre-training Epochs=N/A, Training Paradigm=Supervised Learning Only2024.09 | 68 | — | — | |
| PointNet2025.12 | 68 | — | — | |
| RECONBackbone=Transformer, Ensemble strategy=true, Zero-shot=true2023.02 | 30.5 | — | — | |
| RECONBackbone=Transformer, Ensemble strategy=false, Zero-shot=true2023.02 | 29.5 | — | — | |
| CLIP2PointBackbone=Transformer, Ensemble strategy=true, Zero-shot=true2023.02 | 23.3 | — | — | |
| PointCLIPBackbone=ResNet-50, Ensemble strategy=false, Zero-shot=true2023.02 | 15.4 | — | — | |
| OpenShapePre-Training Dataset=ShapeNet, Evaluation Protocol=Linear Probing2024.02 | — | — | 64.1 | |
| OpenShapePre-Training Dataset=Ensembled, Evaluation Protocol=Linear Probing2024.02 | — | — | 78 | |
| TAMMPre-Training Dataset=ShapeNet, Evaluation Protocol=Linear Probing2024.02 | — | — | 68.5 | |
| TAMMPre-Training Dataset=Ensembled, Evaluation Protocol=Linear Probing2024.02 | — | — | 80.3 | |
| ULIPPre-Training Dataset=ShapeNet, Evaluation Protocol=Linear Probing2024.02 | — | — | 64.8 |