This is an official repository of paper "Combinatorial 3D Shape Generation via Sequential Assembly".

• arXiv Preprint: (abs), (pdf)
• High-Resolution Version (About 50MB)

You are able to install required Python packages by commanding pip install -r requirements.txt.

• Creating a dataset

Run the following script.

# Move to src_dataset/ $ ./dataset_all.sh 

It will create a dataset, which has already been included in the repository.

• Generating a 3D shape

# Move to src_generation/ $ python assemble_with_bo.py --ind_class 21 --ind_target 1 --use_stability --use_rollback

ind_class and ind_target indicate the indices of class and target object, respectively (Please check the code for dataset creation). use_stability and use_rollback are flags for considering stability and using a rollback step.

• Creating an XML file and its corresponding PLY files

Run the following script.

# Move to src_rendering/ $ ./meshes_all.sh 

It requires a rendering process with Mitsuba renderer. After changing the camera position and its perspective, render the XML file you want.

• Bar

• Line

• Plate

• Wall

• Cuboid

• Square Pyramid

• Chair

• Sofa

• Cup

• Hollow

• Table

• Car

@article{KimJ2020arxiv, author={Kim, Jungtaek and Chung, Hyunsoo and Lee, Jinhwi and Cho, Minsu and Park, Jaesik}, title={Combinatorial {3D} Shape Generation via Sequential Assembly}, journal={{arXiv} preprint {arXiv}:2004.07414}, year={2020} } 

or

@inproceedings{KimJ2020neuripsw, author={Kim, Jungtaek and Chung, Hyunsoo and Lee, Jinhwi and Cho, Minsu and Park, Jaesik}, title={Combinatorial {3D} Shape Generation via Sequential Assembly}, booktitle={NeurIPS Workshop on Machine Learning for Engineering Modeling, Simulation, and Design (ML4Eng)}, year={2020} } 

• Jungtaek Kim: jtkim@postech.ac.kr
• Hyunsoo Chung: hschung2@postech.ac.kr

MIT License