SDDPM

<br /> <p align="center"> <h1 align="center">Spiking Denoising Diffusion Probabilistic Models (WACV'24)</h1> <p align="center" > Jiahang Cao^*, Ziqing Wang^*, Hanzhong Guo^*, Hao Cheng, Qiang Zhang, Renjing Xu^† <!-- <a href="https://evelinehong.github.io">Jiahang Cao*</a>, <a href="https://haoyuzhen.com">Ziqing Wang*</a>, <a href="https://peihaochen.github.io">Hanzhong Guo*</a>, <a href="https://zsh2000.github.io">Hao Cheng</a>, <a href="https://yilundu.github.io">Qiang Zhang</a>, <a href="https://zfchenunique.github.io">Renjing Xu</a> --> </p> <!-- <p align="center" > <em>The Hong Kong University of Science and Technology (Guangzhou)</em> </p> --> <p align="center"> <a href='https://arxiv.org/abs/2306.17046'> <img src='https://img.shields.io/badge/Paper-PDF-red?style=flat&logo=arXiv&logoColor=red' alt='Paper PDF'> </a> <a href='https://openaccess.thecvf.com/content/WACV2024/html/Cao_Spiking_Denoising_Diffusion_Probabilistic_Models_WACV_2024_paper.html' style='padding-left: 0.5rem;'> <img src='https://img.shields.io/badge/Proceeding-HTML-blue?style=flat&logo=Google%20chrome&logoColor=blue' alt='Proceeding Supp'> </a> </p> <p align="center"> <img src="figs/illustration_main.png" alt="Logo" width="80%"> </p> </p>

Updates

• 2025.5, we create the discussion section in this repository and provide the insights regarding the recent reproduction of SDDPM (discussion#13). Feel free to start a new discussion or issue!
• 2024.8, we release Spiking Diffusion Model accepted to IEEE Transactions on AI, which (a) extends applicability to a wider array of diffusion solvers, (b) integrates Temporal-wise Spiking Mechanism to capture more dynamic information, and (c) makes the first attempt to utilize an ANN-SNN approach for implementing spiking diffusion models. Codes are available at https://github.com/SageCao1125/SDM!
• 2024.3, the checkpoint of SDDPM in Cifar10 with FID 19 is released!

Requirements

1. Creating conda environment.

conda create -n SDDPM python=3.8
conda activate SDDPM

2. Installing dependencies.

git clone https://github.com/SageCao1125/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.