Home

Awesome

Dynamic Graph CNN for Learning on Point Clouds

We propose a new neural network module dubbed EdgeConv suitable for CNN-based high-level tasks on point clouds including classification and segmentation. EdgeConv is differentiable and can be plugged into existing architectures.

[Project] [Paper] [Press]

Overview

DGCNN is the author's re-implementation of Dynamic Graph CNN, which achieves state-of-the-art performance on point-cloud-related high-level tasks including category classification, semantic segmentation and part segmentation.

<img src='./tensorflow/misc/demo_teaser.png' width=800>

Further information please contact Yue Wang and Yongbin Sun.

Author's Implementations

The classification experiments in our paper are done with the pytorch implementation.

Other Implementations

Generalization under Corruptions

The performance is evaluated on ModelNet-C with mCE (lower is better) and clean OA (higher is better).

MethodReferenceStandalonemCEClean OA
PointNetQi et al.Yes1.4220.907
DGCNNWang et al.Yes1.0000.926

Real-World Applications

Citation

Please cite this paper if you want to use it in your work,

@article{dgcnn,
  title={Dynamic Graph CNN for Learning on Point Clouds},
  author={Wang, Yue and Sun, Yongbin and Liu, Ziwei and Sarma, Sanjay E. and Bronstein, Michael M. and Solomon, Justin M.},
  journal={ACM Transactions on Graphics (TOG)},
  year={2019}
}

License

MIT License

Acknowledgement

The structure of this codebase is borrowed from PointNet.