ShiftDDPMs: Exploring Conditional Diffusion Models by Shifting Diffusion Trajectories

This repository is official PyTorch implementation of ShiftDDPMs (AAAI 2023).

@article{zhang2023shiftddpms,
  title={Shiftddpms: Exploring conditional diffusion models by shifting diffusion trajectories},
  author={Zhang, Zijian and Zhao, Zhou and Yu, Jun and Tian, Qi},
  journal={arXiv preprint arXiv:2302.02373},
  year={2023}
}

Dataset

You can download all datasets from Google Drive .

You should put download under ShiftDDPMs/data/ and unzip them.

Install Requirements

pip install -r requirements.txt

Training

To train regular DDPM, run this command:

bash scripts/dist_train_regular_diffusion.sh 1 0 4

To train ShiftDDPM, run this command:

bash scripts/dist_train_shift_diffusion.sh 1 0 4

You can change the config file and run path in the script file.

Sampling

TODO