Kaibo Tang

**Task** (what are you trying to do/register?) I want to implement `ConditionalTemplateCreation` on PyTorch and test it on my own data of MRI scans of size [160, 192, 160] normalized...

A faster implementation of NCC using cumulative summation

1

Runs 14x faster using a safer approach (to prevent overflow) and 24x faster using an unsafe approach (Tensorflow implementation). Also provide the option to use double precision (to prevent overflow)....

Improved implementation of random flip in `discriminator_augmentations.py`.

Add support for tensor-valued spherical functions in `interp_rbf`

2

Fixes #3236. This PR adds support for tensor-valued spherical functions in `interp_rbf`. When the user calls this function, specifying `legacy=True` internally uses `scipy.interpolate.Rbf` which is the original behavior; specifying `legacy=False`...

Allow `interp_rbf` to accept tensor-valued spherical functions

1

## Description I am currently working with interpolation in the q-space. Let's say I have a series of DWIs of shape `(H, W, L, D)` where `D` is the number...

**Describe the bug** Previously reported [here](https://github.com/Project-MONAI/MONAI/discussions/7289) and [here](https://github.com/Project-MONAI/MONAI/issues/7286). The documentation for `Invert` is [missing](https://docs.monai.io/en/stable/transforms.html#post-processing). **Additional context** It is somewhat confusing to see `Invertd` in the documentation but not `Invert`, especially...

Unexpected Patch2Self behavior when number of diffusion directions is small

4

### Summary The behavior of Patch2Self is unexpected when used to denoise a dataset consisting of a small number of diffusion directions, e.g., 20 b1000s. There are two main issues:...