·
3 commits
to main
since this release
Trained DL models for slice-wise rigid motion correction
This release provides the trained model checkpoints for the moco-dl framework, which performs DL–based slice-wise rigid motion correction in spinal cord dMRI and fMRI. The two model files included in models.zip correspond to separately trained networks for dMRI and fMRI data.
The models were trained using the DenseNet regressor that estimates in-plane translations (Tx and Ty) for each slice and timepoints. When applied with the inference script (test_model.py), the models generate 4D motion-corrected volumes along with the estimated translation fields.
The archive models.zip contains:
dMRI.ckpt – Trained model for dMRI motion correction
fMRI.ckpt – Trained model for fMRI motion correction