Awesome

<a href="https://kangxue.org/LOGAN.html"> LOGAN: Unpaired Shape Transform in Latent Overcomplete Space</a>

<a href="http://kangxue.org">Kangxue Yin</a>, <a href="https://czq142857.github.io/">Zhiqin Chen</a>, <a href="http://vcc.szu.edu.cn/~huihuang/">Hui Huang</a>, <a href="http://www.cs.tau.ac.il/~dcor/">Daniel Cohen-Or</a>, <a href="https://www.cs.sfu.ca/~haoz/">Hao Zhang</a>. <br/>

[<a href="https://arxiv.org/pdf/1903.10170.pdf">Paper</a>] [<a href="https://kangxue.org/papers/logan_supp.pdf">Supplementary material</a>]

Prerequisites

• Linux (tested under Ubuntu 16.04 )
• Python (tested under 3.5.4)
• TensorFlow (tested under 1.12.0-GPU )
• numpy, scipy, etc.

The code is built on the top of <a href="https://github.com/optas/latent_3d_points">latent_3d_points</a> and <a href="https://github.com/charlesq34/pointnet2">pointnet2</a>. Before run the code, please compile the customized TensorFlow operators under the folders "latent_3d_points/structural_losses" and "pointnet_plusplus/tf_ops".

Dataset

• Download the dataset <a href="https://drive.google.com/uc?id=1I0-3UfV7PP5DhTcPvI3BNQv9P1VIhmMF&export=download">HERE</a>.
• If you are in China, you can choose to download it from <a href="https://share.weiyun.com/5WbWcaw"> HERE</a>.

Usage

An example of training and testing the autoencoder:

python -u run_ae.py  --mode=train   --class_name_A=chair    --class_name_B=table    --gpu=0
python -u run_ae.py  --mode=test    --class_name_A=chair    --class_name_B=table    --gpu=0   --load_pre_trained_ae=1

Training and testing the translator:

python -u run_translator.py  --mode=train  --class_name_A=chair    --class_name_B=table   --gpu=0
python -u run_translator.py  --mode=test   --class_name_A=chair    --class_name_B=table   --gpu=0   --load_pre_trained_gan=1

Upsampling:

Hmm.. I haven't put this into the release version of the code. I will work on it as soon as I get time. 

Please note that all the quantitative evaluation results we provided in the paper were done with point clouds of size 2048, i.e., before upsampling.

Citation

If you find our work useful in your research, please consider citing:

@article {yin2019logan,
    author = {Kangxue Yin and Zhiqin Chen and Hui Huang and Daniel Cohen-Or and Hao Zhang}
    title = {LOGAN: Unpaired Shape Transform in Latent Overcomplete Space}
    journal = {ACM Transactions on Graphics(Special Issue of SIGGRAPH Asia)}
    volume = {38}
    number = {6}
    pages = {198:1--198:13}
    year = {2019}
}

Acknowledgments

The code is built on the top of <a href="https://github.com/optas/latent_3d_points">latent_3d_points</a> and <a href="https://github.com/charlesq34/pointnet2">pointnet2</a>. Thanks for the precedent contributions.