Semantic-aware Classifier-Free Guidance (S-CFG)

This is the pytorch implementation of the paper "Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance"

Setup

Our code builds on the reuqirement of the offical Stable Diffusion repository and DeepFloyd IF repository in Hugging Face.

Usage

To generate an image, you can simply run the script by

cd IF
python IF.py

or

cd sd
python sd.py

The detailed settings can be found in the file IF.py and sd.py, where the prompt, and seed can be specified.

Reference

If you find our methods or code helpful, please kindly cite the paper:

@inproceedings{shen2024rethinking,
  title={Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidancee},
  author={Shen, Dazhong and Song, Guanglu and Xue, Zeyue and Wang, Fu-Yun and Liu, Yu},
  booktitle={Proceedings of The IEEE/CVF Computer Vision and Pattern Recognition Conference (CVPR)},
  year={2024}
}