Bayesian Differentiable Physics for Cloth Digitalization

Our CVPR 2024 paper Bayesian Differentiable Physics for Cloth Digitalization.

Needed Compilers and Libraries

• GCC 9.5.0 (or MSVC 19.29.30139)
• CUDA 11.3
• Python 3.8.13
• PyTorch 1.12.1
• Kaolin 0.12.0
• Alglib 3.17.0
• Boost 1.75
• Eigen 3.3.9

How to install

1. Install GCC and CUDA, and confirm their environment variables are set correctly.
2. Install Python (Recommond to use Anaconda).
3. Install Pytorch and Kaolin with following their official documentations.
4. Download Alglib, Boost, and Eigen to a diretory you like.
5. Change the Setup.py to make sure the paths are set correctly, i.e. INCLUDE_DIR.append(...).
6. Run python setup.py install.
7. Finally, you can confirm our simulator has been successfully installed by executing the following commonds in prompt:

-> python
-> import pytorch
-> import diffsim

Check out the python scripts in the folder experiments for training our BDP. They have detailed comments for explaining themselves.

Cusick Drape Dataset

The dataset in given in the folder data.

Authors

Authors Deshan Gong, Ningtao Mao, and He Wang

Deshan Gong, scdg@leeds.ac.uk

He Wang, he_wang@ucl.ac.uk, Personal website

Project Webpage: https://drhewang.com/pages/BDP.html

Citation (Bibtex)

Please cite our paper if you find it useful:

@InProceedings{Gong_Bayesian_2024,
author={Deshan Gong, Ningtao Mao and He Wang},
booktitle={The Conference on Computer Vision and Pattern Recognition (CVPR)},
title={Bayesian Differentiable Physics for Cloth Digitalization},
year={2024}}