This repository contains the official implementation for the paper: Compressible-composable NeRF via Rank-residual Decomposition.

We also provide a slightly different implementation in the torch-ngp framework, which has an interactive GUI and maybe better for experience!

Tested on Ubuntu with Python >= 3.6 and PyTorch >= 1.8.0.

git clone https://github.com/ashawkey/CCNeRF.git
cd CCNeRF
pip install -r requirements.txt 

You can download the following datasets and put them under ./data

• Synthetic-NeRF
• Tanks&Temples

To reproduce the scene in teaser, simply run:

bash run.sh

To generate config files for all objects:

cd configs
# modify the config template in this file.
python gen_config.py 

To train and test on a single object:

# train and test on lego
python train.py --config configs/lego_hybrid.txt
# test with a pretrained checkpoint
python train.py --config configs/lego_hybrid.txt --render_only 1 # choose the default ckpt
python train.py --config configs/lego_hybrid.txt --render_only 1 --ckpt path/to/ckpt # speficy ckpt path

By default, we test and report at all compression levels (groups), which may take some time to finish.

To compose multiple pretrained objects in to a scene, we can modify the composition settings (model checkpoint and transformation matrix) in compose.py. We provide some composed scenes as examples too:

# load model
chair = load_model('./log/chair_hybrid/chair_hybrid_5.th', 'CCNeRF')
# scale and translation
T0 = np.array([
    [0.6, 0, 0, 0.8],
    [0, 0.6, 0, 0],
    [0, 0, 0.6, 0],
    [0, 0, 0, 1],
])
# rotation
R0 = np.eye(4)
R0[:3, :3] = Rot.from_euler('zyx', [-90, 0, 0], degrees=True).as_matrix()
T0 = T0 @ R0
# compose to the scene
tensorf.compose(chair, T0, R0[:3, :3])

The config file is still needed to provide testing camera poses. --ckpt none means we are going to compose on an empty scene, else we will compose on the hotdog scene, which is not desired for the current example.

python compose.py --config configs/hotdog_hybrid.txt --ckpt none

If you find the code useful for your research, please use the following BibTeX entry:

@article{tang2022compressible,
  title={Compressible-composable NeRF via Rank-residual Decomposition},
  author={Tang, Jiaxiang and Chen, Xiaokang and Wang, Jingbo and Zeng, Gang},
  journal={arXiv preprint arXiv:2205.14870},
  year={2022}
}

We would like to thank TensoRF authors for the great framework!