Template-free Articulated Gaussian Splatting for Real-time Reposable Dynamic View Synthesis

🌐Project Page | 🖨️ArXiv | 📰Paper

Install

# install requirements
conda env create -n enviroment.yaml
conda activate SK_GS
# build extension
cd <project root>
cd extenstion/_C
mkdir build
cd build
cmake ..
make -j

Data Prepare

1. Download D-NeRF dataset. Unzip the downloaded data tor prooject root data dir in order to train.
2. Download WIM dataset and Unzip to dir.
3. Prepare ZJU Mocap dataset as watch-it-move
4. Dataset structure

<project root>
├── data
│   ├── DNeRF  
│   │   ├── mutant
│   │   ├── standup 
│   │   ├── ...
│   ├── WIM  
│   │   ├── atlas
│   │   ├── baxter 
│   │   ├── ...
│   ├── zju  
│   │   ├── 313
│   │   ├── ...

Train and Test

python train.py -c exps/d_nerf.yaml --scene hook
python test.py -c exps/d_enrf.yaml --scene hook --load results/DNeRF/last.pth

GUI

python gui.py -c exps/d_enrf.yaml --scene hook --load results/DNeRF/last.pth

Thanks

Thanks to the authors of 3D Gaussians, Deformable-3D-Gaussians and SC-GS for their excellent code.

✏️ Citateion

@InProceedings{SK-GS,
  title = 	 {Template-free Articulated Gaussian Splatting for Real-time Reposable Dynamic View Synthesis},
  author =       {Wan, Diwen and Wang, Yuxiang and Lu, Ruijie and Zeng, Gang},
  booktitle = 	 {NeurIPS},
  year = 	 {2024},
}

@InProceedings{SP-GS,
  title = 	 {Superpoint Gaussian Splatting for Real-Time High-Fidelity Dynamic Scene Reconstruction},
  author =       {Wan, Diwen and Lu, Ruijie and Zeng, Gang},
  booktitle = 	 {Proceedings of the 41st International Conference on Machine Learning},
  pages = 	 {49957--49972},
  year = 	 {2024},
}