LaplacianFusion

This repository contains the accompanying code for LaplacianFusion: Detailed 3D Clothed-Human Body Reconstruction, SIGGRAPH Asia'22

Author Information

• Hyomin Kim
• Hyeonseo Nam
• Jungeon Kim
• Jaesik Park [Google Scholar]
• Seungyong Lee [Google Scholar]

Run LaplacianFusion

Prerequisites

• Ubuntu 18.06 or higher
• CUDA 10.2 or higher
• pytorch 1.9 or higher
• python 3.9 or higher

Download data

• Get sample data and pre-trained 'DVM' weight (111 markers) from here
• Get SMPL-X model from here and make data directory structure as follow:

lapfu
├── dvm_weight.pth
├── human_models
│   └── smplx
│       ├── SMPLX_FEMALE.npz
│       └── SMPLX_MALE.npz
├── protocol_info
└── subjects

• Get SMPL-X code from here and replace "./lib/smplx/" folder

SMPL-X fitting (using Deep Virtual Markers)

• We recommend using docker
• Replace DATADIR in "run_dvm.sh: Line 4" as your path

docker pull min00001/dvm_run
./run_dvm.sh

• You can also use OpenPose feature if color images are given (we already include keypoints and intrinsic parameters in sample dataset)

LaplacianFusion

• Replace DataPath.Main in "config.py: Line 8" as your path

conda create -n lapfu python=3.9
conda activate lapfu
pip install -r ./requirements.txt

python ./preprocessing/fit_smplx.py
./script/run_learning.sh

• You can find result meshes in "lapfu/subjects/*/train/recon"

License

This software is being made available under the terms in the LICENSE file.

Any exemptions to these terms requires a license from the Pohang University of Science and Technology.

Citing LaplacianFusion

@inproceedings{Kim_LaplacianFusion_SIGGRAPH_Asia_2022,
Title={LaplacianFusion: Detailed 3D Clothed-Human Body Reconstruction},
Author={Hyomin Kim and Hyeonseo Nam and Jungeon Kim and Jaesik Park and Seungyong Lee},
Booktitle={Proceedings of the ACM (SIGGRAPH Asia)},
Year={2022}
}