Official implementation of "Low-dose CT Denoising with Language-engaged Dual-space Alignment" arxiv
March, 2024: initial commit.
The 2016 AAPM-Mayo dataset can be downloaded from: CT Clinical Innovation Center (B30 kernel)
The 2020 AAPM-Mayo dataset can be downloaded from: cancer imaging archive
Mayo2016_2d/
|--train/
|--quarter_1mm/
train_quarter_00001.npy
train_quarter_00002.npy
train_quarter_00003.npy
...
|--full_1mm/
train_full_00001.npy
train_full_00002.npy
train_full_00003.npy
...
|--test/
|--quarter_1mm
|--full_1mm
- Linux Platform
- torch==1.12.1+cu113 # depends on the CUDA version of your machine
- torchvision==0.13.1+cu113
- Python==3.8.0
- numpy==1.22.3
Build per-layer candidate token:python process_words.py
Then we used the official repository of VQ-GAN (https://github.com/CompVis/taming-transformers) to set up training. Please refer to (models/taming) to learn about our modifications to original VQ-GAN.
python train.py --name LEDA(experiment_name) --model LEDA --netG redcnn --dataroot /data/zhchen/Mayo2016_2d(path to images) --lr 0.0002 --gpu_ids 6,7 --print_freq 25 --batch_size 8 --lr_policy cosine
python test.py --name LEDA(experiment_name) --model LEDA --netG redcnn --results_dir test_results --result_name LEDA_results(path to save image) --gpu_ids 6 --batch_size 1 --eval
Please refer to options files for more setting.
If you find our work and code helpful, please kindly cite the corresponding paper:
@article{chen2024low,
title={Low-dose CT Denoising with Language-engaged Dual-space Alignment},
author={Chen, Zhihao and Chen, Tao and Wang, Chenhui and Niu, Chuang and Wang, Ge and Shan, Hongming},
journal={arXiv preprint arXiv:2403.06128},
year={2024}
}