DIRECT — Calgary-Campinas model zoo
Pretrained multi-coil MRI reconstruction baselines for the
Calgary-Campinas MR reconstruction challenge,
repackaged as <name>_<Rx>.yaml + <name>_<Rx>.pt pairs for DIRECT.
Originally hosted at files.aiforoncology.nl/direct-project. Metrics and context: DIRECT model zoo.
Dataset
| Dataset | Calgary-Campinas brain MRI |
| Challenge | MR reconstruction challenge |
| Training | 47 multi-coil volumes (12 coils), retrospective 5× / 10× undersampling |
| Validation | 20 volumes |
| Masks | Challenge Poisson-disk masks — also on Hub as NKI-AI/direct-mri-masks |
Models
Filenames encode the fixed challenge acceleration (5x or 10x):
rim_5x.yaml / rim_5x.pt
rim_10x.yaml / rim_10x.pt
varnet_5x.yaml / …
Architectures include RIM, VarNet, LPDNet, XPDNet, KIKI-Net, U-Net, RecurrentVarNet, ConjGradNet, IterDualNet, JointICNet, and others. YAML configs match projects/calgary_campinas/ in the DIRECT repository.
Training / inference masking
Each checkpoint was trained for a single challenge rate (5× or 10×) with the CalgaryCampinas mask. Inference YAMLs already pin that one acceleration. To run the other rate, use the matching *_5x / *_10x file pair.
Install DIRECT
git clone https://github.com/NKI-AI/direct.git
cd direct
conda create --name direct python=3.12
conda activate direct
pip install meson-python meson ninja
pip install --no-build-isolation -e ".[dev]"
Usage
hf download NKI-AI/direct-calgary-campinas --local-dir ./calgary
direct predict ./predictions \
--cfg ./calgary/rim_5x.yaml \
--checkpoint ./calgary/rim_5x.pt \
--data-root /path/to/calgary_campinas \
--num-gpus 1
The first argument is the prediction output directory.
License
Creative Commons Attribution-ShareAlike 3.0 (aligned with the DIRECT model zoo release).
Citation
If you use these models or DIRECT, please cite:
@article{DIRECTTOOLKIT,
title={DIRECT: Deep Image REConstruction Toolkit},
author={Yiasemis, George and Moriakov, Nikita and Karkalousos, Dimitrios and Caan, Matthan and Teuwen, Jonas},
journal={Journal of Open Source Software},
volume={7},
number={73},
pages={4278},
year={2022},
doi={10.21105/joss.04278},
url={https://doi.org/10.21105/joss.04278}
}