chrockey/atlasnetv2

This repository contains the source codes for the paper AtlasNet V2 - Learning Elementary Structures.

AtlasNet V2 - Learning Elementary Structures

This work was build upon Thibault Groueix's AtlasNet and 3D-CODED projects. (you might want to have a look at those)

This repository contains the source codes for the paper AtlasNet V2 - Learning Elementary Structures.

Citing this work

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

@inproceedings{deprelle2019learning,
  title={Learning elementary structures for 3D shape generation and matching},
  author={Deprelle, Theo and Groueix, Thibault and Fisher, Matthew and Kim, Vladimir and Russell, Bryan and Aubry, Mathieu},
  booktitle={Advances in Neural Information Processing Systems},
  pages={7433--7443},
  year={2019}
}

Project Page

The project page is available http://imagine.enpc.fr/~deprellt/atlasnet2/

Install

Clone the repo and install dependencies

This implementation uses Pytorch.

## Download the repository
git clone https://github.com/TheoDEPRELLE/AtlasNetV2.git
cd AtlasNetV2
## Create python env with relevant packages
conda create --name atlasnetV2 python=3.7
source activate atlasnetV2
pip install pandas visdom
conda install pytorch torchvision -c pytorch
conda install -c conda-forge matplotlib
# you're done ! Congrats :)

Training

Data

cd data; ./download_data.sh; cd ..

We used the ShapeNet dataset for 3D models.

When using the provided data make sure to respect the shapenet license.

• The point clouds from ShapeNet, with normals go in data/customShapeNet
• The corresponding normalized mesh (for the metro distance) goes in data/ShapeNetCorev2Normalized
• the rendered views go in data/ShapeNetRendering

The trained models and some corresponding results are also available online :

• The trained_models go in trained_models/

Build chamfer distance

The chamfer loss is based on a custom cuda code that need to be compile.

source activate pytorch-atlasnet
cd ./extension
python setup.py install

Start training

• First launch a visdom server :

python -m visdom.server -p 8888

• Check out all the options :

git pull; python training/train.py --help

• Run the baseline :

git pull; python training/train.py --model AtlasNet --adjust mlp
git pull; python training/train.py --model AtlasNet --adjust linear

• Run the Patch Deformation module with the different adjustment modules :

git pull; python training/train.py --model PatchDeformation --adjust mlp
git pull; python training/train.py --model PatchDeformation --adjust linear

• Run the Point Translation module with the different adjustment modules:

git pull; python training/train.py --model PointTranslation --adjust mlp
git pull; python training/train.py --model PointTranslation --adjust linear

• Monitor your training on http://localhost:8888/

Models

The models train on the SURREAL dataset for the FAUST competition can be found here

Acknowledgement

This work was partly supported by ANR project EnHerit ANR-17-CE23-0008, Labex Bezout, and gifts from Adobe to Ecole des Ponts.

License

MIT