Awesome

Compressing Volumetric Radiance Fields to 1 MB (CVPR2023)

<a href="https://arxiv.org/abs/2211.16386"><img src="https://img.shields.io/badge/arXiv-2211.16386-b31b1b.svg" height=22.5></a>

Lingzhi Li*, Zhen Shen*, Zhongshu Wang, Li Shen, Liefeng Bo

Alibaba Group

:+1:Update: We Released VQ-TensoRF :+1:

Citation:

@inproceedings{li2023compressing,
  title={Compressing volumetric radiance fields to 1 mb},
  author={Li, Lingzhi and Shen, Zhen and Wang, Zhongshu and Shen, Li and Bo, Liefeng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={4222--4231},
  year={2023}
}

Note: This repository only contain VQ-DVGO.

Setup

• Download datasets: NeRF NSVF, T&T (masked)

• Install required libraries

pip install -r requirements.txt
cd lib/cuda
python setup.py install

Please install the correct version of Pytorch and torch_scatter for your machine.

Directory structure for the datasets

data
├── nerf_synthetic     # Link: https://drive.google.com/drive/folders/128yBriW1IG_3NJ5Rp7APSTZsJqdJdfc1
│   └── [chair|drums|ficus|hotdog|lego|materials|mic|ship]
│       ├── [train|val|test]
│       │   └── r_*.png
│       └── transforms_[train|val|test].json
│
├── Synthetic_NSVF     # Link: https://dl.fbaipublicfiles.com/nsvf/dataset/Synthetic_NSVF.zip
│   └── [Bike|Lifestyle|Palace|Robot|Spaceship|Steamtrain|Toad|Wineholder]
│       ├── intrinsics.txt
│       ├── rgb
│       │   └── [0_train|1_val|2_test]_*.png
│       └── pose
│           └── [0_train|1_val|2_test]_*.txt
│
│
└── TanksAndTemple     # Link: https://dl.fbaipublicfiles.com/nsvf/dataset/TanksAndTemple.zip
    └── [Barn|Caterpillar|Family|Ignatius|Truck]
        ├── intrinsics.txt
        ├── rgb
        │   └── [0|1|2]_*.png
        └── pose
            └── [0|1|2]_*.txt
    

Training

# for nerf_synthetic datasets
python autotask_final.py -g "0 1 2 3 4 5 6 7"  --configname syn_4096code 

# for Synthetic_NSVF datasets
python autotask_final.py -g "0 1 2 3 4 5 6 7"  --configname nsvf_4096code --dataset nsvf

# for TanksAndTemple datasets
python autotask_final.py -g "0 1 2 3 4 5 6 7"  --configname tnt_4096code  --dataset tnt

Fixing TanksandTemples dataset following this issue

Set -g option according to the availible gpu on your machine.

Testing

# for nerf_synthetic datasets
python autotask_eval_only.py -g "0 1 2 3 4 5 6 7"  --configname syn_4096code 

# for Synthetic_NSVF datasets
python autotask_eval_only.py -g "0 1 2 3 4 5 6 7"  --configname nsvf_4096code --dataset nsvf

# for TanksAndTemple datasets
python autotask_eval_only.py -g "0 1 2 3 4 5 6 7"  --configname tnt_4096code  --dataset tnt