Object Classification on ModelNet40
94.3Instance AccuracyUtonia
Evaluation results
| Method | Links | ||
|---|---|---|---|
| UtoniaParams Learn.=137.4M, Params Pct.=100%, Evaluation Protocol=fine-tuning2026.03 | 94.3 | 92.4 | |
| SonataParams Learn.=124.8M, Params Pct.=100%, Evaluation Protocol=fine-tuning2026.03 | 94.1 | 92.1 | |
| ConcertoParams Learn.=137.4M, Params Pct.=100%, Evaluation Protocol=fine-tuning2026.03 | 94.1 | 92.4 | |
| MKConvOperation=Conv., Input=xyz, #points=1024, voting strategy=true2021.07 | 94 | — | |
| Point-M2AEParams Learn.=12.9M, Params Pct.=100%, Evaluation Protocol=fine-tuning2026.03 | 94 | — | |
| PAConvOperation=Conv., Input=xyz, #points=1024, voting strategy=true2021.07 | 93.9 | — | |
| PointTrans.Operation=Trans., Input=xyz, #points=10242021.07 | 93.7 | — | |
| MKConvOperation=Conv., Input=xyz, #points=1024, voting strategy=false2021.07 | 93.7 | — | |
| RS-CNNOperation=Conv., Input=xyz, #points=1024, voting strategy=true2021.07 | 93.6 | — | |
| PAConvOperation=Conv., Input=xyz, #points=1024, voting strategy=false2021.07 | 93.6 | — | |
| PatchFormerOperation=Trans., Input=xyz, #points=10242021.07 | 93.5 | — | |
| SO-NetOperation=MLP, Input=xyz, nr, #points=50002021.07 | 93.4 | — | |
| DensePointOperation=MLP, Input=xyz, #points=10242021.07 | 93.2 | — | |
| ShellNetOperation=MLP, Input=xyz, #points=10242021.07 | 93.1 | — | |
| InterpCNNOperation=Conv., Input=xyz, #points=10242021.07 | 93 | — | |
| KPConvOperation=Conv., Input=xyz, #points=68002021.07 | 92.9 | — | |
| RS-CNNOperation=Conv., Input=xyz, #points=1024, voting strategy=false2021.07 | 92.9 | — | |
| PointASNLOperation=Conv., Input=xyz, #points=10242021.07 | 92.9 | — | |
| PTv3Params Learn.=124.8M, Params Pct.=100%2026.03 | 92.9 | 90.1 | |
| PointConvOperation=Conv., Input=xyz, nr, #points=10242021.07 | 92.5 | — | |
| FPConvOperation=Conv., Input=xyz, #points=10242021.07 | 92.5 | — | |
| DenX-ConvOperation=Conv., Input=xyz, #points=10242021.07 | 92.5 | — | |
| SpiderCNNOperation=Conv., Input=xyz, nr, #points=10242021.07 | 92.4 | — | |
| PointCNNOperation=Conv., Input=xyz, #points=10242021.07 | 92.2 | — | |
| DGCNNOperation=MLP, Input=xyz, #points=10242021.07 | 92.2 | — | |
| UtoniaParams Learn.=<0.2M, Params Pct.=<0.2%, Evaluation Protocol=linear probing2026.03 | 90.8 | 88.2 | |
| PointNet++Operation=MLP, Input=xyz, #points=10242021.07 | 90.7 | — | |
| ConcertoParams Learn.=<0.2M, Params Pct.=<0.2%, Evaluation Protocol=linear probing2026.03 | 90.7 | 88.2 | |
| SonataParams Learn.=<0.2M, Params Pct.=<0.2%, Evaluation Protocol=linear probing2026.03 | 89.8 | 85.1 | |
| PointNetOperation=MLP, Input=xyz, #points=10242021.07 | 89.2 | — | |
| SO-Net (3-layer)Representation=points + normal, Input=5000 x 6, Training time=3h, Pre-training=false2018.03 | 0.934 | 0.908 | |
| SO-Net (2-layer)Representation=points + normal, Input=5000 x 6, Training time=3h, Pre-training=true2018.03 | 0.925 | 0.894 | |
| SO-Net (2-layer)Representation=points + normal, Input=5000 x 6, Training time=3h, Pre-training=false2018.03 | 0.923 | 0.893 | |
| PointNet++Representation=points + normal, Input=5000 x 6, Training time=20h2018.03 | 0.919 | — | |
| Kd-NetRepresentation=points, Input=2^15 x 3, Training time=120h2018.03 | 0.918 | 0.885 | |
| SO-Net (2-layer)Representation=points, Input=2048 x 3, Training time=3h, Pre-training=false2018.03 | 0.909 | 0.873 | |
| O-CNNRepresentation=octree, Input=64^32018.03 | 0.906 | — | |
| DeepSetsRepresentation=points, Input=5000 x 32018.03 | 0.9 | — | |
| PointNetRepresentation=points, Input=1024 x 3, Training time=3-6h2018.03 | 0.892 | 0.862 | |
| ECCRepresentation=points, Input=1000 x 32018.03 | 0.874 | 0.832 | |
| OctNetRepresentation=octree, Input=128^32018.03 | 0.865 | 0.838 |