ReDiffusion

July 25, 2023 ยท View on GitHub

The code for the ICML 2023 paper ReDi: Efficient Learning-Free Diffusion Inference via Trajectory Retrieval.

Installation

  1. Install diffusers from source.
git clone https://github.com/huggingface/diffusers.git
cd diffusers
git checkout 25f1142
pip install -e .
  1. Put the pipeline_re_sd.py in ./src/diffusers/pipelines and import it in ./src/diffusers/__init__.py.

Knowledge Base Creation

  1. Follow this (instruction)[https://github.com/facebookresearch/faiss/blob/main/INSTALL.md] to install faiss-cpu.

  2. Run traj.py to construct the knowledge base.

Image generation

  1. retrieve_val10.py gives an example of generating images with ReDi one by one.

  2. retrieve_redi.py gives an example of generating images with ReDi in batches.