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.
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.
-
Use
h5_dataloader.py
download andload modelnet40_ply_hdf5_2048
files -
Custom down-sampling points from mesh. Download Modelnet40 off file, and unzip it in
Data/ModelNet40
RunSampler
withtest = 0
andtest = 1
, and sampled point-cloud file will save inModelNet40_
Next runpointcloud_dataloader
to convert*.points
to h5 file.Data\ModelNet40_
folder will createModelNet40_test.h5
andModelNet40_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