3D Object Classification on Sydney Few-shot learning setup sparse (100 points)
86.25-way 10-shot AccuracyOur+DGCNN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Our+DGCNNBackbone=DGCNN, Pre-training=Our self-supervised point embeddings2020.09 | 86.2 | 90.9 | 66.15 | 81.5 | |
| PointNet++Evaluation Protocol=FSL setting, Supervision=Supervised (Random Weight Initialization)2020.09 | 79.89 | 84.99 | 55.35 | 63.35 | |
| Our+PointNetBackbone=PointNet, Pre-training=Our self-supervised point embeddings2020.09 | 76.5 | 83.7 | 55.45 | 64 | |
| PointCNNEvaluation Protocol=FSL setting, Supervision=Supervised (Random Weight Initialization)2020.09 | 75.83 | 83.43 | 56.27 | 73.05 | |
| PointNetEvaluation Protocol=FSL setting, Supervision=Supervised (Random Weight Initialization)2020.09 | 74.16 | 82.18 | 51.35 | 58.3 | |
| Latent-GANEvaluation Protocol=Linear SVM classification on final embeddings, Supervision=Unsupervised2020.09 | 64.5 | 79.8 | 50.45 | 62.5 | |
| PointCapsNetEvaluation Protocol=Linear SVM classification on final embeddings, Supervision=Unsupervised2020.09 | 59.44 | 70.5 | 44.1 | 60.25 | |
| FoldingNetEvaluation Protocol=Linear SVM classification on final embeddings, Supervision=Unsupervised2020.09 | 58.9 | 71.2 | 42.6 | 63.45 | |
| DGCNNEvaluation Protocol=FSL setting, Supervision=Supervised (Random Weight Initialization)2020.09 | 58.3 | 76.7 | 48.05 | 76.1 | |
| 3D-GANEvaluation Protocol=Linear SVM classification on final embeddings, Supervision=Unsupervised2020.09 | 54.2 | 58.8 | 36 | 45.25 | |
| Voxel+DGCNNBackbone=DGCNN, Pre-training=VoxelSSL [26]2020.09 | 52.5 | 79.6 | 52.65 | 69.1 |