Home

Awesome

AdaDiff

Official PyTorch implementation of AdaDiff described in the paper.

Salman UH Dar, Şaban Öztürk, Yilmaz Korkmaz, Gokberk Elmas, Muzaffer Özbey, Alper Güngör, Tolga Çukur, "Adaptive Diffusion Priors for Accelerated MRI Reconstruction", arXiv 2022.

<img src="./figures/main_fig.png" width="600px"> <img src="./figures/main_fig_opt.png" width="600px">

Dependencies

python>=3.6.9
torch>=1.7.1
torchvision>=0.8.2
cuda=>11.2
ninja
python3.x-dev (apt install, x should match your python3 version, ex: 3.8)

Installation

git clone https://github.com/icon-lab/AdaDiff
cd AdaDiff

Train

<br />

For Single-Coil

python3 train_adadiff_singlecoil.py --dataset brain --image_size 256 --exp experiment_name_for_singlecoil --num_channels 1 --num_channels_dae 64 --ch_mult 1 1 2 2 4 4 --num_timesteps 8 --num_res_blocks 2 --batch_size 8 --num_epoch 500 --ngf 64 --embedding_type positional --use_ema --ema_decay 0.999 --r1_gamma 1. --z_emb_dim 256 --lr_d 1e-4 --lr_g 1.6e-4 --lazy_reg 10 --num_process_per_node 1 --save_content --local_rank 0

For Multi-Coil

python3 train_adadiff_multicoil.py --dataset brain_multi_coil --image_size 288 --exp experiment_name_for_multicoil --num_channels 2 --num_channels_dae 64 --ch_mult 1 1 1 2 2 --num_timesteps 8 --num_res_blocks 2 --batch_size 4 --num_epoch 500 --ngf 64 --embedding_type positional --use_ema --ema_decay 0.999 --r1_gamma 1. --z_emb_dim 256 --lr_d 1e-4 --lr_g 1.6e-4 --lazy_reg 10 --num_process_per_node 1 --save_content --attn_resolutions 18 --local_rank 0
<br />

Inference

<br />

For Single-Coil

python3 inference_adadiff_singlecoil.py --dataset brain --image_size 256 --exp content_folder_for_singlecoil --num_channels 1 --num_channels_dae 64 --ch_mult 1 1 2 2 4 4 --num_timesteps 8 --num_res_blocks 2 --epoch_id 1000 --batch_size 1 --lr_g 1e-3 --itr_inf 1000 --local_rank 0 --contrast T1 --phase test --save_inter True --R 4 --extra_string lr_1e3_opt_reset_ --reset_opt True

For Multi-Coil

python3 inference_adadiff_multicoil.py --dataset brain_multi_coil --image_size 288 --exp content_folder_for_multicoil --num_channels 2 --num_channels_dae 64 --ch_mult 1 1 1 2 2 --num_timesteps 8 --num_res_blocks 2  --epoch_id 1000 --batch_size 1 --lr_g 1e-3 --itr_inf 1000 --attn_resolutions 18 --local_rank 0 --phase test --contrast T1 --R 4 --save_inter True --extra_string lr_1e3_opt_reset --reset_opt True
<br /> <br />

Citation

You are encouraged to modify/distribute this code. However, please acknowledge this code and cite the paper appropriately.

@article{dar2022adaptive,
  title={Adaptive Diffusion Priors for Accelerated MRI Reconstruction},
  author={Dar, Salman UH and {\"O}zt{\"u}rk, {\c{S}}aban and Korkmaz, Yilmaz and Elmas, Gokberk and {\"O}zbey, Muzaffer and G{\"u}ng{\"o}r, Alper and {\c{C}}ukur, Tolga},
  journal={arXiv preprint arXiv:2207.05876},
  year={2022}
}

(c) ICON Lab 2022

<br />

Acknowledgements

This code uses libraries from, pGAN, StyleGAN-2, and DD-GAN repositories.