GitHub - l1346792580123/FastHuman: [3DV 2024] FastHuman: Reconstructing High-Quality Clothed Human in Minutes. (original) (raw)

wild.mp4

Installation

clone the repository

git clone https://github.com/l1346792580123/FastHuman.git cd FastHuman

Step 1: requirements:

pip install -r requirements.txt

Step 2: install PyTorch (The PyTorch version should be higher than 1.7.1.):

pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

Step 3: install nvdiffrast

git clone https://github.com/NVlabs/nvdiffrast cd nvdiffrast pip install .

Data Preparation

You can download NHR data and DTU data from NHR and DTU respectively.

Run

When you have installed the environment and downloaded the data. You need to change the data path of the conigs files. Then you can run the code.

python space_carving.py --conf confs/nhr_sp.conf --scan_id 1 python ncc_optim.py --conf confs/nhr_ncc.conf --scan_id 1 python sfs_optim.py --conf confs/nhr_sfs.conf --scan_id 1

space_carving.py generates the initial mesh. ncc_optim.py employs multi-view patch-based photometric optimization. sfs_optim.py applies shape from shading refinement.

Reconstruction Results of NHR dataset

ret3.mp4 ret4.mp4

Citation

@inproceedings{fasthuman,
  author={Lin, Lixiang, Peng Songyou, Gan Qijun and Zhu, Jianke},
  booktitle={International Conference on 3D Vision, 3DV}, 
  title={FastHuman: Reconstructing High-Quality Clothed Human in Minutes}, 
  year={2024},
  }

Multiview Textured Mesh Recovery by Differentiable Rendering