deepali
deepali copied to clipboard
BioMedIA
Image, point set, and surface registration in PyTorch.
It has been noted by some collaborators that the MI implementation ([here](https://github.com/BioMedIA/deepali/blob/372b943b007a43fd573ba02b4b4b5ad1868907e8/src/deepali/losses/functional.py#L975)) in this library is slightly slower and less easy to tune than De Vos's implementation in TorchIR (https://github.com/BDdeVos/TorchIR/blob/main/torchir/metrics.py#L74)....
The functions in [deepali.core._kornia](https://github.com/BioMedIA/deepali/blob/main/src/deepali/core/_kornia.py) can be replaced by the functionality available in the new and lightweight [RoMa](https://naver.github.io/roma/) library.
Currently, most flow field functions assume that the number of vector field channels is equal to the number of grid dimensions. However, in case of a spatio-temporal flow field, each...
Started to extend the tutorial to include a learning based image registration example. It is not currently producing any meaningful results.
Currently, a `SpatialTransform` is a coordinate mapping of normalized coordinates with respect to the reference sampling `Grid` specified when constructing the spatial transformation object. A user of the transform has...
Mask in ncc_loss from deepali.losses.functional does not work. the image-loss is already collapses to shape=[c] before masked_loss is called, through the sum(dim=1) calls.
Hi! `import deepali.core.functional as U from .config import join_kwargs_in_sequence ` These imports are missing in the newest version from the `/home/pti/anaconda3/envs/reg10/lib/python3.10/site-packages/deepali/core/__init__.py` and as a result the ffd example is not...
https://github.com/BioMedIA/deepali/blob/ae3e4762a20cbbd0711cc5dd53a7f1aafad49b86/src/deepali/utils/imageio/nifti.py#L116 The Nifti saving function using nibabel here does not work because the affine matrix is incomplete. The output affine matrix from `grid.affine()` is 3x3 which is missing the origin/translation...
