[CVPR2024] Adversarial Score Distillation: When score distillation meets GAN

Arxiv | Paper | Project page

Overview (see Project page for more examples)

Generated 3D NeRFs

Generated 3D Gaussians

Installation

Install ASD requirements, Differential Gaussian Rasterization and simple-knn

git clone https://github.com/2y7c3/ASD
cd ASD

### for 3D Gaussian Splatting
#git clone --recursive https://github.com/YixunLiang/diff-gaussian-rasterization
#pip install ./diff-gaussian-rasterization

#git clone https://github.com/YixunLiang/simple-knn.git
#pip install ./simple-knn

pip install -r requirements.txt

Optional

pip install ninja

Install tiny-cuda-nn

For 3D Gaussian Splatting

Install Point-E

Install Shape-E

Download finetuned Shap-E by Cap3D, and put it in ./load

Quick start

# NeRF Training
python launch.py --config configs/test_nerf.yaml --train --gpu 0 system.prompt_processor.prompt="A delicious hamburger"

# 3D gaussian Training (experimental implementation)
#python launch.py --config configs/test_gs.yaml --train --gpu 0 system.prompt_processor.prompt="A delicious hamburger"

# NeRF Tuning
# you might want to resume training from the certain checkpoint
python launch.py --config configs/test_tune_nerf.yaml --train --gpu 0 system.prompt_processor.prompt="A delicious hamburger" resume="path/to/ckpt"

# NeRF Testing 
# you can change camera parameters on here
python launch.py --config configs/test_tune_nerf.yaml --test --gpu 0 system.prompt_processor.prompt="A delicious hamburger" resume="path/to/ckpt"

Todo

[x] Release the training codes for NeRF
[-] Release the training codes for 3DGS
[ ] Release the training codes for 2D images and image editing

Citation

If you find our work useful in your research, please consider citing:

@InProceedings{Wei_2024_CVPR,
    author    = {Wei, Min and Zhou, Jingkai and Sun, Junyao and Zhang, Xuesong},
    title     = {Adversarial Score Distillation: When score distillation meets GAN},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024},
    pages     = {8131-8141}
}

Acknowledgements

This code is built on many research works and open-source projects:

Thanks for their excellent works.

2y7c3 / ASD

readme