Point Cloud Classification on ModelNet40 (test)
94.7AccuracyRepSurf-U
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RepSurf-UPre-train=N/A, Tr. Param.=1.5 M2022.08 | 94.7 | — | — | — | — | — | |
| ReconInput=PC & I & L, Cross-Modal Transfer=Contrastive Learning, Masking Strategies=Global Random, #Params (M)=140.9, GFLOPS=20.9, Times (h)=34, Evaluation Protocol=Fine-tuned2023.12 | 94.5 | — | — | — | — | — | |
| Point-FEMAEInput=PC, Masking Strategies=Hybrid Global & Local, #Params (M)=41.5, GFLOPS=5.0, Times (h)=21, Evaluation Protocol=Fine-tuned2023.12 | 94.5 | — | — | — | — | — | |
| Ours (C+S)Voting=true2023.06 | 94.2 | — | — | — | — | — | |
| PointMLPPre-train=N/A, Tr. Param.=12.6 M2022.08 | 94.1 | — | — | — | — | — | |
| I2PInput=PC & I, Cross-Modal Transfer=Projection & 2D Recon., Masking Strategies=2D-Guided, #Params (M)=74.9, GFLOPS=16.8, Times (h)=64, Evaluation Protocol=Fine-tuned2023.12 | 94.1 | — | — | — | — | — | |
| PointMLPApproach=Hierarchical, point grouping2024.02 | 94.1 | — | — | — | — | — | |
| PointNeXtPre-train=N/A, Tr. Param.=1.4 M2022.08 | 94 | — | — | — | — | — | |
| P2PPre-train=2D, Tr. Param.=1.2 M, Backbone=HorNet-L-22k-mlp2022.08 | 94 | — | — | — | — | — | |
| Point-M2AEInput=PC, Masking Strategies=Multi-Scale Global Random, #Params (M)=15.3, GFLOPS=3.7, Times (h)=29, Evaluation Protocol=Fine-tuned2023.12 | 94 | — | — | — | — | — | |
| Joint-MAEInput=PC & I, Cross-Modal Transfer=Projection & 2D Recon., Masking Strategies=Global Random, Evaluation Protocol=Fine-tuned2023.12 | 94 | — | — | — | — | — | |
| Point-M2AEVoting=true2023.06 | 94 | — | — | — | — | — | |
| Point-MAEPre-train=3D, Tr. Param.=21.1 M2022.08 | 93.8 | — | — | — | — | — | |
| Point-MAEInput=PC, Masking Strategies=Global Random, #Params (M)=29.0, GFLOPS=2.3, Times (h)=13, Evaluation Protocol=Fine-tuned2023.12 | 93.8 | — | — | — | — | — | |
| MaskPointVoting=true2023.06 | 93.8 | — | — | — | — | — | |
| Point-MAEVoting=true2023.06 | 93.8 | — | — | — | — | — | |
| CLoRAPre-trained model=RECON [62], Fine-tuning strategy=CLoRA(ours), #Param. (M)=0.4 (1.81%), #GFLOPs=4.76(-)2025.12 | 93.8 | — | — | — | — | — | |
| ACTInput=PC, Cross-Modal Transfer=Knowledge Distillation, Masking Strategies=Global Random, #Params (M)=135.5, GFLOPS=31.0, Times (h)=52, Evaluation Protocol=Fine-tuned2023.12 | 93.7 | — | — | — | — | — | |
| Ours (C+S)Voting=false2023.06 | 93.7 | — | — | — | — | — | |
| PointMLP-elitePre-train=N/A, Tr. Param.=0.68 M2022.08 | 93.6 | — | — | — | — | — | |
| Ours (S)Voting=false2023.06 | 93.6 | — | — | — | — | — | |
| CLoRAPre-trained model=Point-MAE [61], Fine-tuning strategy=CLoRA(ours), #Param. (M)=0.3 (1.36%), #GFLOPs=4.76(-)2025.12 | 93.6 | — | — | — | — | — | |
| PointGSTPre-trained model=RECON [62], Fine-tuning strategy=PointGST [56], #Param. (M)=0.6 (2.77%), #GFLOPs=4.81(↑)2025.12 | 93.6 | — | — | — | — | — | |
| DAPTPre-trained model=Point-MAE [61], Fine-tuning strategy=DAPT [41], #Param. (M)=1.1 (4.97%), #GFLOPs=4.96(↑)2025.12 | 93.5 | — | — | — | — | — | |
| PointGSTPre-trained model=Point-MAE [61], Fine-tuning strategy=PointGST [56], #Param. (M)=0.6 (2.77%), #GFLOPs=4.81(↑)2025.12 | 93.5 | — | — | — | — | — | |
| DAPTPre-trained model=RECON [62], Fine-tuning strategy=DAPT [41], #Param. (M)=1.1 (4.97%), #GFLOPs=4.96(↑)2025.12 | 93.5 | — | — | — | — | — | |
| PCT-3LVoting=false2023.06 | 93.4 | — | — | — | — | — | |
| Point-M2AEVoting=false2023.06 | 93.4 | — | — | — | — | — | |
| Point-PEFTPre-trained model=Point-BERT [60], Fine-tuning strategy=Point-PEFT [63], #Param. (M)=0.7 (3.13%), #GFLOPs=7.61(↑)2025.12 | 93.4 | — | — | — | — | — | |
| PointGSTPre-trained model=Point-BERT [60], Fine-tuning strategy=PointGST [56], #Param. (M)=0.6 (2.77%), #GFLOPs=4.81(↑)2025.12 | 93.4 | — | — | — | — | — | |
| CLoRAPre-trained model=Point-BERT [60], Fine-tuning strategy=CLoRA(ours), #Param. (M)=0.3 (1.36%), #GFLOPs=4.76(-)2025.12 | 93.4 | — | — | — | — | — | |
| IDPTPre-trained model=RECON [62], Fine-tuning strategy=IDPT [55], #Param. (M)=1.7 (7.69%), #GFLOPs=7.10(↑)2025.12 | 93.4 | — | — | — | — | — | |
| PointLoRAPre-trained model=RECON [62], Fine-tuning strategy=PointLoRA [64], #Param. (M)=0.8 (3.43%), #GFLOPs=5.06(↑)2025.12 | 93.4 | — | — | — | — | — | |
| IDPTPre-trained model=Point-MAE [61], Fine-tuning strategy=IDPT [55], #Param. (M)=1.7 (7.69%), #GFLOPs=7.10(↑)2025.12 | 93.3 | — | — | — | — | — | |
| Point-PEFTPre-trained model=Point-MAE [61], Fine-tuning strategy=Point-PEFT [63], #Param. (M)=0.7 (3.13%), #GFLOPs=7.61(↑)2025.12 | 93.3 | — | — | — | — | — | |
| PointLoRAPre-trained model=Point-MAE [61], Fine-tuning strategy=PointLoRA [64], #Param. (M)=0.8 (3.43%), #GFLOPs=5.06(↑)2025.12 | 93.3 | — | — | — | — | — | |
| Point-PEFTPre-trained model=RECON [62], Fine-tuning strategy=Point-PEFT [63], #Param. (M)=0.7 (3.13%), #GFLOPs=7.61(↑)2025.12 | 93.3 | — | — | — | — | — | |
| Point-BERTPre-train=3D, Tr. Param.=21.1 M2022.08 | 93.2 | — | — | — | — | — | |
| Point-BERTVoting=true2023.06 | 93.2 | — | — | — | — | — | |
| Point-MAEVoting=false2023.06 | 93.2 | — | — | — | — | — | |
| PCT-2LVoting=false2023.06 | 93.2 | — | — | — | — | — | |
| FullPre-trained model=Point-MAE [61], Fine-tuning strategy=Full, #Param. (M)=22.1 (100%), #GFLOPs=4.762025.12 | 93.2 | — | — | — | — | — | |
| PointLoRAPre-trained model=Point-BERT [60], Fine-tuning strategy=PointLoRA [64], #Param. (M)=0.9 (4.07%), #GFLOPs=5.06(↑)2025.12 | 93.2 | — | — | — | — | — | |
| FG-Net2020.12 | 93.1 | 0.0561 | — | — | — | — | |
| P2PPre-train=2D, Tr. Param.=0.25 M, Backbone=ResNet-1012022.08 | 93.1 | — | — | — | — | — | |
| BiXTApproach=Point grouping & hierarchy2024.02 | 93.1 | — | — | — | — | — | |
| DAPTPre-trained model=Point-BERT [60], Fine-tuning strategy=DAPT [41], #Param. (M)=1.1 (4.97%), #GFLOPs=4.96(↑)2025.12 | 93.1 | — | — | — | — | — | |
| DGCNN-OCCoPre-train=3D, Tr. Param.=1.8 M2022.08 | 93 | — | — | — | — | — | |
| KPConv2020.11 | 92.9 | — | — | — | — | — | |
| KPConvPre-train=N/A, Tr. Param.=15.2 M2022.08 | 92.9 | — | — | — | — | — | |
| DGCNNPre-train=N/A, Tr. Param.=1.8 M2022.08 | 92.9 | — | — | — | — | — | |
| DGCNNMode=Device-Only (D), Device=Jetson TX2, Edge=None2025.12 | 92.9 | 0.2419 | — | 88.9 | 1 | — | |
| DGCNNMode=Device-Only (D), Device=Raspberry Pi 4B, Edge=None2025.12 | 92.9 | 1.1218 | — | 88.9 | 5.6 | — | |
| DGCNNMode=Edge-Only (E), Edge=Nvidia GPU, Device=Jetson TX2, Network Bandwidth (S_L)=<= 40 Mbps2025.12 | 92.9 | 0.1188 | — | 88.9 | 0.2 | — | |
| DGCNNMode=Edge-Only (E), Edge=Nvidia GPU, Device=Jetson TX2, Network Bandwidth (S_L)=<= 10 Mbps2025.12 | 92.9 | 0.1239 | — | 88.9 | 0.3 | — | |
| Point Transformer2020.11 | 92.8 | — | — | — | — | — | |
| Point-BERTVoting=false2023.06 | 92.7 | — | — | — | — | — | |
| FullPre-trained model=Point-BERT [60], Fine-tuning strategy=Full, #Param. (M)=22.1 (100%), #GFLOPs=4.762025.12 | 92.7 | — | — | — | — | — | |
| Point2Sequence2020.11 | 92.6 | — | — | — | — | — | |
| IDPTPre-trained model=Point-BERT [60], Fine-tuning strategy=IDPT [55], #Param. (M)=1.7 (7.69%), #GFLOPs=7.10(↑)2025.12 | 92.6 | — | — | — | — | — | |
| [10]Mode=Device-Only (D), Device=Jetson TX2, Edge=None2025.12 | 92.6 | 0.1076 | — | 90.6 | 0.4 | — | |
| [10]Mode=Device-Only (D), Device=Raspberry Pi 4B, Edge=None2025.12 | 92.6 | 0.8511 | — | 90.6 | 4.3 | — | |
| BiXTApproach=Point grouping2024.02 | 92.5 | — | — | — | — | — | |
| FullPre-trained model=RECON [62], Fine-tuning strategy=Full, #Param. (M)=22.1 (100%), #GFLOPs=4.762025.12 | 92.5 | — | — | — | — | — | |
| PointCNN2020.11 | 92.2 | — | — | — | — | — | |
| HGNASMode=Edge-Only (E), Edge=Intel CPU, Device=Jetson TX2, Network Bandwidth (S_L)=<= 10 Mbps2025.12 | 92.1 | 0.0883 | — | 88.5 | 0.2 | — | |
| BranchyMode=Co-Inference (Co), Edge=Intel CPU, Device=Jetson TX2, Network Bandwidth (S_L)=<= 40 Mbps2025.12 | 92 | 0.1402 | — | — | 0.6 | — | |
| PointNet++Geometric features=extra, Augmentation=advanced2021.03 | 91.9 | — | — | — | — | — | |
| PointNet++2020.11 | 91.9 | — | — | — | — | — | |
| PAT2020.11 | 91.7 | — | — | — | — | — | |
| PWavePBackbone=PointNet, Attack=Point Addition2026.02 | 91.06 | — | — | — | — | — | |
| APESM=512, Pre-processing=true2025.04 | 90.81 | — | — | — | — | — | |
| PointNet++Pre-train=N/A, Tr. Param.=1.4 M2022.08 | 90.7 | — | — | — | — | — | |
| PointNet++Approach=Hierarchical, point grouping2024.02 | 90.7 | — | — | — | — | — | |
| SAMBLEM=5122025.04 | 90.58 | — | — | — | — | — | |
| PointNet++Scheme=Clean2024.10 | 90.55 | — | — | — | — | — | |
| SpiderCNN2020.11 | 90.5 | — | — | — | — | — | |
| Set Transformer2020.11 | 90.4 | — | — | — | — | — | |
| APESM=256, Pre-processing=true2025.04 | 90.4 | — | — | — | — | — | |
| Deep Sets2020.11 | 90.3 | — | — | — | — | — | |
| SAMBLEM=2562025.04 | 90.18 | — | — | — | — | — | |
| SAMBLEM=1282025.04 | 90.02 | — | — | — | — | — | |
| LighTNM=5122025.04 | 89.91 | — | — | — | — | — | |
| APESM=512, Pre-processing=false2025.04 | 89.81 | — | — | — | — | — | |
| SAMBLEM=642025.04 | 89.81 | — | — | — | — | — | |
| ShapeContextNet2020.11 | 89.8 | — | — | — | — | — | |
| APESM=128, Pre-processing=true2025.04 | 89.77 | — | — | — | — | — | |
| BiXT (naïve)Approach=Naïve, point-based2024.02 | 89.6 | — | — | — | — | — | |
| APESM=64, Pre-processing=true2025.04 | 89.57 | — | — | — | — | — | |
| DGCNNScheme=Clean2024.10 | 89.51 | — | — | — | — | — | |
| SAMBLEM=322025.04 | 89.45 | — | — | — | — | — | |
| PointNet2020.11 | 89.2 | — | — | — | — | — | |
| PointNetApproach=Naïve, point-based2024.02 | 89.2 | — | — | — | — | — | |
| DA-NetM=5122025.04 | 89.01 | — | — | — | — | — | |
| PointConvScheme=Clean2024.10 | 88.9 | — | — | — | — | — | |
| PointCVARBackbone=PointNet, Attack=Point Addition2026.02 | 88.57 | — | — | — | — | — | |
| APESM=32, Pre-processing=true2025.04 | 88.56 | — | — | — | — | — | |
| FPSM=5122025.04 | 88.34 | — | — | — | — | — | |
| LighTNM=2562025.04 | 88.21 | — | — | — | — | — | |
| SampleNetM=5122025.04 | 88.16 | — | — | — | — | — |