Animatable Neural Radiance Fields from Monocular RGB Videos

Github | Paper

Animatable Neural Radiance Fields from Monocular RGB Videos
Jianchuan Chen, Ying Zhang, Di Kang, Xuefei Zhe, Linchao Bao, Xu Jia, Huchuan Lu

Demos

male-3-casual

More demos please see Demos.

Requirements

Python3.7 or later
PyTorch 1.6 or later
Pytorch-lightning
KNN_CUDA

For visualization

pyrender
Trimesh
PyMCubes

Run the following code to install all pip packages:

pip install -r requirements.txt

To install KNN_CUDA, we provide two ways:

from source

git clone https://github.com/unlimblue/KNN_CUDA.git
cd KNN_CUDA
make && make install

from wheel

pip install --upgrade https://github.com/unlimblue/KNN_CUDA/releases/download/0.2/KNN_CUDA-0.2-py3-none-any.whl

SMPL models

To download the SMPL model go to this (male, female and neutral models).

Place them as following:

smplx
└── models
    └── smpl
        ├── SMPL_FEMALE.pkl
        ├── SMPL_MALE.pkl
        └── SMPL_NEUTRAL.pkl

Data Preparation

People-Snapshot datasets

prepare images and smpls

python -m tools.people_snapshot --data_root ${path_to_people_snapshot_datasets} --people_ID male-3-casual --gender male --output_dir data/

prepare template

python -m tools.prepare_template --data_root data/ people_ID male-3-casual --model_type smpl --gender male --model_path ${path_to_smpl_models}

Training

Training on the training frames

python train.py --cfg_file configs/people_snapshot/male-3-casual.yml

Finetuning the smpl params on the testing frames

python train.py --cfg_file configs/people_snapshot/male-3-casual_refine.yml train.ckpt_path checkpoints/male-3-casual/last.ckpt

We provide the preprocessed data and pretrained models at Here

Visualization

Novel view synthesis

python novel_view.py --ckpt_path checkpoints/male-3-casual/last.ckpt

3D reconstruction

python extract_mesh.py --ckpt_path checkpoints/male-3-casual/last.ckpt

Novel pose synthesis

python novel_pose.py --ckpt_path checkpoints/male-3-casual/last.ckpt

Testing

python test.py --ckpt_path checkpoints/male-3-casual_refine/last.ckpt --vis

Citation

If you find the code useful, please cite:

@misc{chen2021animatable,
      title={Animatable Neural Radiance Fields from Monocular RGB Videos}, 
      author={Jianchuan Chen and Ying Zhang and Di Kang and Xuefei Zhe and Linchao Bao and Xu Jia and Huchuan Lu},
      year={2021},
      eprint={2106.13629},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Acknowledgements

Parts of the code were based on from kwea123's NeRF implementation: https://github.com/kwea123/nerf_pl. Some functions are borrowed from PixelNeRF https://github.com/sxyu/pixel-nerf

woo1 / Anim-NeRF