Kaiqi Wang, Ke Chen*, Kui Jia     IJCAI 2019

[paper] | [project page]

This implementation uses Pytorch.

git clone https://github.com/wkqscut/DCGNet.git
cd DCGNet
## Create python env with relevant packages
conda create --name dcg python=3.7
conda activate dcg
pip install -U pip
pip install -r requirements.txt
conda install pytorch torchvision cudatoolkit=9.0 -c pytorch  # cudatoolkit=10.0 for cuda10

Tested on pytorch >= 1.0 and python3.

## Build chamfer distance
conda activate dcg
cd ./extension
python setup.py install
cd ../

We used the rendered imaged from 3d-R2N2, and the groundtruth 3D point clouds sampled from ShapeNet.

• The rendered images (*.png) go in ./data/ShapeNetRendering
• The ShapeNetPoints (*.mat) datasets go in ./data/ShapeNetPoints

• The pretrained models (*.pth) should be placed in ./trained_models/

## unzip the Pretrained models using the scripts
bash ./trained_models/unzip_models_dataset.sh

• demo code for DCGNet

bash ./scripts/demo.sh

Make sure that the visdom is alive before training:

python -m visdom.server -p 8990 (change the port if in use)

• train the DCGNet for Point Set AutoEncoding:

bash ./scripts/train_svr_dcg.sh

• train the DCGNet for Point Set Reconstruction from a Single Image:

bash ./scripts/train_svr_dcg.sh

• test the DCGNet for Point Set AutoEncoding:

bash ./scripts/test_svr_dcg.sh

• test the DCGNet for Point Set Reconstruction from a Single Image:

bash ./scripts/test_svr_dcg.sh

If you find this code useful for your research, please consider citing the following paper:

@inproceedings{ijcai2019-517,
  title     = {Deep Cascade Generation on Point Sets},
  author    = {Wang, Kaiqi and Chen, Ke and Jia, Kui},
  booktitle = {Proceedings of the Twenty-Eighth International Joint Conference on
               Artificial Intelligence, {IJCAI-19}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  pages     = {3726--3732},
  year      = {2019},
  month     = {7},
  doi       = {10.24963/ijcai.2019/517},
  url       = {https://doi.org/10.24963/ijcai.2019/517},
}

This work is supported in part by the Program for Guangdong Introducing Innovative and Enterpreneurial Teams (Grant No.: 2017ZT07X183), the National Natural Science Foundation of China (Grant No.: 61771201), and the Program of the Construction of Talented Personnel by the South China University of Technology (Grant No.: D6192110).