- Python 3.8
- PyTorch == 1.13.1
- Transformers == 4.25.1
- diffusers == 0.8.0
- NumPy
pip install -r requirements.txt
We give 7 datasets to evaluate your editing behavior. Each pair has reference image and target prompt. Datasets are from https://github.com/SiatMMLab/Awesome-Diffusion-Model-Based-Image-Editing-Methods?tab=readme-ov-file#benchmark-editeval_v1.
To change edit datasets, just change --input_path
and --data_path
. And the output structure is below:
output
|-ref_image
|-1.jpg
|-2.jpg
|-tar_image
|-1.jpg
|-2.jpg