MMPS Benchmark

This fork of the official DPS repository is adapted to benchmark diffusion posterior sampling algorithms, including DPS, PiGDM, TMPD and DiffPIR against MMPS.

Getting started

1. Clone the repository and its dependencies

   git clone https://github.com/francois-rozet/mmps-benchmark
   cd mmps-benchmark
   git clone https://github.com/VinAIResearch/blur-kernel-space-exploring bkse
   git clone https://github.com/LeviBorodenko/motionblur motionblur
   

2. Install the Python dependencies

   pip install torch==2.3.1 torchvision==0.18.1 --index-url https://download.pytorch.org/whl/cu121
   pip install -r requirements.txt
   

3. Download a checkpoint from link and place it into ./models

   mkdir models
   mv ffhq_10m.pt ./models
   

4. Run an experiment

   python run.py \
   --model-config ./configs/model_ffhq_config.yaml \
   --task-config ./configs/tasks/inpainting_random_config.yaml \
   --method mmps --steps 100 --maxiter 5 \
   --seed 42 \
   

   The generated images are saved in /results and the metrics are written in metrics.csv.