Home

Awesome

Deep Diffusion Image Prior for Efficient OOD Adaptation in 3D Inverse Problems

Hyungjin Chung and Jong Chul Ye

📢📢 The pre-trained model checkpoints and data are moved to a new location.

Official PyTorch implementation for Deep Diffusion Image Prior (DDIP), presented in the paper Deep Diffusion Image Prior for Efficient OOD Adaptation in 3D Inverse Problems, ECCV 2024. problem_setting concept main_results

Getting Started

Download pre-trained modei weights

# Download pre-trained model weights
mkdir -p './exp/vp'
wget -O './exp/vp/ellipses_ema.pth' 'https://www.dropbox.com/scl/fi/g2yd0ecpboz3vc8iwc530/ellipses_ema.pt?rlkey=i1yff5lj2ynk6of0uywqxnqyv&st=smhuffir&dl=1'
wget -O './exp/vp/fastmri_brain_320_complex_1m.pth' 'https://www.dropbox.com/scl/fi/febne0udjvq0cphbggrmy/fastmri_brain_320_complex_1m.pt?rlkey=k4e8tk21ueqjslsw17b05sbho&st=tain0kew&dl=1'

Download sample test data

# Download sample test data
mkdir -p './data'
wget -O './data/data.zip' 'https://www.dropbox.com/scl/fo/wqgpu59c1ge5gw6uwgldw/AHkRLVMkeyr-Odo4CbNtRYI?rlkey=mo6dcglz9pcsjinvgjvby5bm7&st=6n9okudo&dl=1'
# Extract zip file
unzip -q ./data/data.zip -d ./data

By default, the above scripts places the pre-trained model checkpoints under exp/vp, and the sample data under data.

Inverse Problem Solving

Each experiment in the paper can be reproduced by simply running the scripts in ./scripts. All scripts will run by going through the main.py file.

Citation

If you find our work interesting, please consider citing

@article{chung2024deep,
  title={Deep Diffusion Image Prior for Efficient OOD Adaptation in 3D Inverse Problems},
  author={Chung, Hyungjin and Ye, Jong Chul},
  journal={arXiv preprint arXiv:2407.10641},
  year={2024}
}