Few-shot 3D Object Classification (5-way) on ModelNet40 (test)
9810-shot AccuracyPointGPT-L
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PointGPT-LModel Size=Large, Learning Type=Self-Supervised Representation Learning, Larger Pre-training=true, Post-pre-training stage excluded=true2023.05 | 98 | 99 | — | |
| PointGPT-BModel Size=Base, Learning Type=Self-Supervised Representation Learning, Larger Pre-training=true, Post-pre-training stage excluded=true2023.05 | 97.5 | 98.8 | — | |
| RECONTransfer Protocol=MLP-3, Backbone=RECON-block, Voting Strategy=False2023.02 | 97.4 | 98.5 | — | |
| RECONReference=ICML 23, Tuning Paradigm=Self-Supervised Representation Learning (FULL), Context=baseline2024.08 | 97.3 | 98.9 | — | |
| ReConLearning Type=Cross-modal information and teacher models2023.05 | 97.3 | 98.9 | — | |
| RECONTransfer Protocol=FULL, Backbone=RECON-block, Voting Strategy=False2023.02 | 97.3 | 98.9 | — | |
| I2P-MAEReference=CVPR 23, Tuning Paradigm=Self-Supervised Representation Learning (FULL)2024.08 | 97 | 98.3 | — | |
| RECON w/ PPTReference=ACMMM 25, Tuning Paradigm=Parameter-Efficient Supervised Fine-tuning2024.08 | 97 | 98.7 | — | |
| VPPReference=NeurIPS 23, Tuning Paradigm=Self-Supervised Representation Learning (FULL)2024.08 | 96.9 | 98.3 | — | |
| RECON w/ IDPTReference=ICCV 23, Tuning Paradigm=Parameter-Efficient Supervised Fine-tuning, reproduced=true2024.08 | 96.9 | 98.3 | — | |
| RECONTransfer Protocol=MLP-LINEAR, Backbone=RECON-block, Voting Strategy=False2023.02 | 96.9 | 98.2 | — | |
| Point-M2AEReference=NeurIPS 22, Tuning Paradigm=Self-Supervised Representation Learning (FULL)2024.08 | 96.8 | 98.3 | — | |
| ACTReference=ICLR 23, Tuning Paradigm=Self-Supervised Representation Learning (FULL)2024.08 | 96.8 | 98 | — | |
| Point-M2AELearning Type=Self-Supervised Representation Learning2023.05 | 96.8 | 98.3 | — | |
| PointGPT-SModel Size=Small, Learning Type=Self-Supervised Representation Learning, Larger Pre-training=false, Post-pre-training stage excluded=false2023.05 | 96.8 | 98.6 | — | |
| ACTLearning Type=Cross-modal information and teacher models2023.05 | 96.8 | 98 | — | |
| Point-M2AETransfer Protocol=FULL, Backbone=Transformer, Voting Strategy=False2023.02 | 96.8 | 98.3 | — | |
| ACTTransfer Protocol=FULL, Backbone=Transformer, Voting Strategy=False2023.02 | 96.8 | 98 | — | |
| Point-MAE†Transfer Protocol=FULL, Backbone=RECON-block, Voting Strategy=False2023.02 | 96.4 | 97.8 | — | |
| Point-MAEReference=ECCV 22, Tuning Paradigm=Self-Supervised Representation Learning (FULL)2024.08 | 96.3 | 97.8 | — | |
| Point-MAELearning Type=Self-Supervised Representation Learning2023.05 | 96.3 | 97.8 | — | |
| Point-MAETransfer Protocol=FULL, Backbone=Transformer, Voting Strategy=False2023.02 | 96.3 | 97.8 | — | |
| ACTTransfer Protocol=MLP-3, Backbone=Transformer, Voting Strategy=False2023.02 | 95.9 | 97.7 | — | |
| RECON w/ DAPTReference=CVPR 24, Tuning Paradigm=Parameter-Efficient Supervised Fine-tuning, reproduced=true2024.08 | 95.6 | 97.7 | — | |
| MaskPointLearning Type=Self-Supervised Representation Learning2023.05 | 95 | 97.2 | — | |
| MaskPointTransfer Protocol=FULL, Backbone=Transformer, Voting Strategy=False2023.02 | 95 | 97.2 | — | |
| Point-MAETransfer Protocol=MLP-3, Backbone=Transformer, Voting Strategy=False2023.02 | 95 | 96.7 | — | |
| Point-BERTReference=CVPR 22, Tuning Paradigm=Self-Supervised Representation Learning (FULL)2024.08 | 94.6 | 96.3 | — | |
| Point-BERTLearning Type=Self-Supervised Representation Learning2023.05 | 94.6 | 96.3 | — | |
| Point-BERTTransfer Protocol=FULL, Backbone=Transformer, Voting Strategy=False2023.02 | 94.6 | 96.3 | — | |
| OcCoReference=ICCV 21, Tuning Paradigm=Self-Supervised Representation Learning (FULL)2024.08 | 94 | 95.9 | — | |
| OcCoTransfer Protocol=FULL, Backbone=Transformer, Voting Strategy=False2023.02 | 94 | 95.9 | — | |
| ACTTransfer Protocol=MLP-LINEAR, Backbone=Transformer, Voting Strategy=False2023.02 | 91.8 | 93.1 | — | |
| Point-MAETransfer Protocol=MLP-LINEAR, Backbone=Transformer, Voting Strategy=False2023.02 | 91.1 | 91.7 | — | |
| OcCoReference=ICCV 21, Tuning Paradigm=Supervised2024.08 | 90.6 | 92.5 | — | |
| OcCoLearning Type=Supervised2023.05 | 90.6 | 92.5 | — | |
| OcCoTransfer Protocol=None, Backbone=Transformer, Voting Strategy=False2023.02 | 90.6 | 92.5 | — | |
| Transformer†Transfer Protocol=FULL, Backbone=RECON-block, Voting Strategy=False2023.02 | 90.2 | 94.3 | — | |
| TransformerTransfer Protocol=FULL, Backbone=Transformer, Voting Strategy=False2023.02 | 87.8 | 93.3 | — | |
| GPr-Net (Hyp)Num Points=512, #Params=1.24K, F/B pass=50KB, Distance Metric=Hyperbolic2023.04 | 81.13 | 82.71 | — | |
| GPr-Net (Hyp)Num Points=1024, #Params=1.24K, F/B pass=50KB, Distance Metric=Hyperbolic2023.04 | 80.4 | 81.99 | — | |
| Enrich-FeaturesNum Points=10242023.04 | 76.69 | 85.76 | — | |
| GPr-Net (Euc)Num Points=1024, #Params=1.24K, F/B pass=50KB, Distance Metric=Euclidean2023.04 | 74.37 | 75.12 | — | |
| GPr-Net (Euc)Num Points=512, #Params=1.24K, F/B pass=50KB, Distance Metric=Euclidean2023.04 | 74.04 | 74.98 | — | |
| PointCNNNum Points=10242023.04 | 65.41 | 68.64 | — | |
| SS-FSL (PointNet)Num Points=1024, #Params=3.47M, F/B pass=8.5GB2023.04 | 63.2 | 68.9 | — | |
| SS-FSL (DGCNN)Num Points=1024, #Params=1.82M, F/B pass=53GB2023.04 | 60 | 65.7 | — | |
| 3D-GANNum Points=10242023.04 | 55.8 | 65.8 | — | |
| PointNetNum Points=1024, #Params=3.47M, F/B pass=8.5GB2023.04 | 51.97 | 57.81 | — | |
| PointCapsNetNum Points=1024, #Params=2.15M, F/B pass=39GB2023.04 | 42.3 | 53 | — | |
| Latent-GANNum Points=10242023.04 | 41.6 | 46.2 | — | |
| PointNet++Num Points=1024, #Params=1.48M, F/B pass=149GB2023.04 | 38.53 | 42.39 | — | |
| FoldingNetNum Points=1024, #Params=0.67M, F/B pass=5.7GB2023.04 | 33.4 | 35.8 | — | |
| DGCNNReference=TOG 19, Tuning Paradigm=Supervised2024.08 | 31.6 | 40.8 | — | |
| DGCNNNum Points=1024, #Params=1.82M, F/B pass=53GB2023.04 | 31.6 | 40.8 | — | |
| DGCNNLearning Type=Supervised2023.05 | 31.6 | 40.8 | — | |
| DGCNNTransfer Protocol=None, Backbone=DGCNN, Voting Strategy=False2023.02 | 31.6 | 40.8 | — | |
| ACT (baseline)Backbone=ACT, Voting Strategy=without voting2024.10 | — | — | 98 | |
| ACT + DAPTBackbone=ACT, Voting Strategy=without voting2024.10 | — | — | 97.3 | |
| ACT + IDPTBackbone=ACT, Voting Strategy=without voting2024.10 | — | — | 98.2 | |
| ACT + Point-PEFTBackbone=ACT, Voting Strategy=without voting2024.10 | — | — | 97.7 | |
| ACT + PointGSTBackbone=ACT, Voting Strategy=without voting2024.10 | — | — | 98.6 | |
| Point-BERT (baseline)Backbone=Point-BERT, Voting Strategy=without voting2024.10 | — | — | 96.3 | |
| Point-BERT + DAPTBackbone=Point-BERT, Voting Strategy=without voting2024.10 | — | — | 97.3 | |
| Point-BERT + IDPTBackbone=Point-BERT, Voting Strategy=without voting2024.10 | — | — | 97.2 | |
| Point-BERT + Point-PEFTBackbone=Point-BERT, Voting Strategy=without voting2024.10 | — | — | 97.3 | |
| Point-BERT + PointGSTBackbone=Point-BERT, Voting Strategy=without voting2024.10 | — | — | 97.9 | |
| Point-MAE (baseline)Backbone=Point-MAE, Voting Strategy=without voting2024.10 | — | — | 97.8 | |
| Point-MAE + DAPTBackbone=Point-MAE, Voting Strategy=without voting2024.10 | — | — | 98 | |
| Point-MAE + IDPTBackbone=Point-MAE, Voting Strategy=without voting2024.10 | — | — | 97.9 | |
| Point-MAE + Point-PEFTBackbone=Point-MAE, Voting Strategy=without voting2024.10 | — | — | 97.6 | |
| Point-MAE + PointGSTBackbone=Point-MAE, Voting Strategy=without voting2024.10 | — | — | 98.3 | |
| PointGPT-L (baseline)Backbone=PointGPT-L, Voting Strategy=without voting2024.10 | — | — | 99 | |
| PointGPT-L + DAPTBackbone=PointGPT-L, Voting Strategy=without voting2024.10 | — | — | 98.4 | |
| PointGPT-L + IDPTBackbone=PointGPT-L, Voting Strategy=without voting2024.10 | — | — | 98.3 | |
| PointGPT-L + Point-PEFTBackbone=PointGPT-L, Voting Strategy=without voting2024.10 | — | — | 98 | |
| PointGPT-L + PointGSTBackbone=PointGPT-L, Voting Strategy=without voting2024.10 | — | — | 99.6 | |
| RECON (baseline)Backbone=RECON, Voting Strategy=without voting2024.10 | — | — | 98.9 | |
| RECON + DAPTBackbone=RECON, Voting Strategy=without voting2024.10 | — | — | 97.3 | |
| RECON + IDPTBackbone=RECON, Voting Strategy=without voting2024.10 | — | — | 98.6 | |
| RECON + Point-PEFTBackbone=RECON, Voting Strategy=without voting2024.10 | — | — | 97.7 | |
| RECON + PointGSTBackbone=RECON, Voting Strategy=without voting2024.10 | — | — | 98.7 |