A Re-Implement of Dynamic Graph CNN for Point-Cloud Classification and Segmentation

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]

DGCNN-Pytorch is my personal re-implementation of Dynamic Graph CNN.

There is two ways to convert ModelNet40 PLY or OFF file to Point-Cloud.

1. Use h5_dataloader.py download and load modelnet40_ply_hdf5_2048 files

2. Custom down-sampling points from mesh. Download Modelnet40 off file, and unzip it in Data/ModelNet40 Run Sampler with test = 0 and test = 1, and sampled point-cloud file will save in ModelNet40_ Next run pointcloud_dataloader to convert *.points to h5 file. Data\ModelNet40_ folder will create ModelNet40_test.h5 and ModelNet40_train.h5

Train model: Run train to train your model. Now is PointNet, next will update to pointnet and DGCNN.

Next few days, will upload DGCNN model.

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}
}

MIT License