MIR.models.MultiMorph

MultiMorph group-wise registration model for on-demand atlas construction.

Abulnaga, S. M., Hoopes, A., Dey, N., Hoffmann, M., Fischl, B., Guttag, J., & Dalca, A. (2025). MultiMorph: On-demand Atlas Construction. In Proceedings of the Computer Vision and Pattern Recognition Conference (pp. 30906-30917).

Code retrieved from: https://github.com/mabulnaga/multimorph

Modified and tested by: Junyu Chen jchen245@jhmi.edu Johns Hopkins University

Functions

make_epoch_batches(n_samples, min_bs, max_bs)

Create randomized batch index lists for one epoch.

Classes

ComposeCentrality(*args, **kwargs)	Compose group-wise fields with a shared central field.
DeformationFieldComposer(*args, **kwargs)	Compose group-wise displacement fields.
FastMeanConv3d(*args, **kwargs)	Perform a group mean convolution (see UniverSeg paper).
FastMeanConv3dUp(*args, **kwargs)	Perform a group mean convolution for the upsampling UNet.
GroupConv2d(*args, **kwargs)	Perform a group convolution without communication.
GroupConv3d(*args, **kwargs)	Perform a group convolution without communication.
GroupNet(*args, **kwargs)	MultiMorph 2D group U-Net predicting per-entry displacement fields.
GroupNet3D(*args, **kwargs)	MultiMorph 3D group U-Net predicting per-entry diffeomorphic fields.
Lambda(*args, **kwargs)	Wrap a callable inside an nn.Module.
ListBatchSampler(*args, **kwargs)	Yield precomputed lists of indices as batches.
MaxPool2d(*args, **kwargs)	2D max pooling applied to each group entry of a [B, N, C, H, W] tensor.
MaxPool3d(*args, **kwargs)	3D max pooling applied to each group entry of a [B, N, C, H, W, D] tensor.
MeanConv2d(*args, **kwargs)	Perform a group mean convolution (see UniverSeg paper).
MeanConv3d(*args, **kwargs)	Perform a group mean convolution (see UniverSeg paper).
ResizeTransform(*args, **kwargs)	Resize and rescale a vector field transform.
SimpleUNet(*args, **kwargs)	Plain 2D U-Net baseline (no group dimension).
SubtractMean(*args, **kwargs)	Subtract mean along a specified dimension.
UpsamplingBilinear2d(*args, **kwargs)	Bilinear upsampling applied to each group entry of a [B, N, C, H, W] tensor.
UpsamplingTrilinear3d(*args, **kwargs)	Trilinear upsampling applied to each group entry of a [B, N, C, H, W, D] tensor.
VecInt(*args, **kwargs)	Integrate a vector field via scaling and squaring.
VecIntGroup(*args, **kwargs)	Vector integration with group dimension.
Warp2d(*args, **kwargs)	Warp each group entry of a 2D group tensor with its own displacement field.
Warp3d(*args, **kwargs)	Warp each group entry of a 3D group tensor with its own displacement field.