- paper link:https://arxiv.org/pdf/1801.07829.pdf
- Author's code repo:WangYueFt/dgcnn (github.com). The code has both included pytorch & tensorflow versions.
- Dataset:ModelNet40
- 12,311 meshed CAD models
- 40 categories
- 9,843 models are used for training and 2,468 models are for testing.
refer to ModelNet40
| Dataset | Paper | Paper(avg) | Our(tf) | Our(tf,avg) | Our(th) | Our(th,avg) |
|---|---|---|---|---|---|---|
| ModelNet40 | 91.7 | 88.9 | 88.94 | 83.94 | 88.01 | 82.16 |
Notes: accuracy with 'avg' refers to 'average accuracy', which calls the method of balanced_accuracy_score; accuracy without 'avg' refers to 'overall accuracy', which calls the method of accuracy_score.