Spiking Denoising Diffusion Probabilistic Models (WACV'24)

Jiahang Cao^*, Ziqing Wang^*, Hanzhong Guo^*, Hao Cheng, Qiang Zhang, Renjing Xu^†

Requirements

1. (Optional) Creating conda environment.

conda create -n SDDPM python=3.8
conda activate SDDPM

2. Installing dependencies.

git clone https://github.com/AndyCao1125/SDDPM.git
cd SDDPM
pip install -r requirements.txt

Training

Codes for training Spiking Denoising Diffusion Probabilistic models.

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python main_SDDPM.py \
    --train \
    --dataset='cifar10' \
    --beta_1=1e-4 --beta_T=0.02 \
    --img_size=32 --timestep=4 --img_ch=3 \
    --parallel=True --sample_step=0 \
    --total_steps=500001 \
    --logdir='./logs' \
    --wandb

Evaluation

Codes for evaluating the image generation quantitative results, including FID and IS score.

[Update March.19th] The checkpoint of SDDPM in CIFAR-10 is now released. You can download the checkpoint through this link.

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python main_SDDPM.py \
    --eval \
    --pre_trained_path 'your/model' \
    --num_images 50000 \        
    --fid_cache './stats/cifar10.train.npz'

Image Generation Results

Citation

If you find our work useful, please consider citing:

@inproceedings{cao2024spiking,
  title={Spiking denoising diffusion probabilistic models},
  author={Cao, Jiahang and Wang, Ziqing and Guo, Hanzhong and Cheng, Hao and Zhang, Qiang and Xu, Renjing},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={4912--4921},
  year={2024}
}

Acknowledgements & Contact

We thank the authors (pytorch-ddpm, spikingjelly) for their open-sourced codes.

For any help or issues of this project, please contact jcao248@connect.hkust-gz.edu.cn.