The official repository for the paper "A Three-player GAN for Super-Resolution in Magnetics Resonance Imaging" in MLCN workshop of MICCAI2023
- Patch the entire brain volume into smaller ones, to be compatible with GPU memories;
- Training on the patches
- Inferring the test data
- Assembling testing patches
python crop_nifti_9t.py <your data folder path>or in multiprocessing way:
python mp_crop_nifti.py <your data folder path>python ./mains/ln_DDP_train.py --model_name 'ThreePlayerGAN'this loads the configure YAML file in ./config folder, of course you can write your own config file or even the training script.
python ./mains/inference/inference_WholeBrain.py [argvs] # skipping patching and assembling, memory-UNfriendly, but you can trade-off it with speed by placing them on CPU
# or
python ./mains/inference/inference.py [argvs] # also including pathcing and assembling, but trivial difference between stiched patches existplease refer to utils README for detailed inference introduction.
It mainly serves as an utility module for the inference steps, mainly stored in ./mains/inference/assemble_einops.py
@InProceedings{threeplayergan,
author="Wang, Qi and Mahler, Lucas and Steiglechner, Julius and Birk, Florian and Scheffler, Klaus and Lohmann, Gabriele",
title="A Three-Player GAN for Super-Resolution in Magnetic Resonance Imaging",
booktitle="Machine Learning in Clinical Neuroimaging",
year="2023",
publisher="Springer Nature Switzerland",
pages="23--33",
isbn="978-3-031-44858-4"
}