batch SimilarityTransform

The following code implements face alignment using functions from the `skimage` library. In cases where there are a small number of faces, using this function for face alignment can yield satisfactory results. However, when dealing with a large collection of faces, I'm looking for a method to calculate similarity transformation matrices in batches. I have already achieved batch face alignment using similarity transformation matrices of shape [b, 2, 3] with the `kornia` library. However, I'm struggling to find a way to calculate similarity transformation matrices in batches while maintaining consistent results with the computation performed by `skimage`. Furthermore, I'm hoping to accomplish this using the PyTorch framework. I have attempted to replicate the computation of similarity transformation matrices from `skimage` using PyTorch, but the results do not match. This discrepancy could impact the accuracy of subsequent face recognition tasks. Has anyone successfully implemented this? Any help would be greatly appreciated.

```
tform = trans.SimilarityTransform()
tform.estimate(src,dst)
M = tform.params[0:2,:]
```
can stack the results of `trans.SimilarityTransform()` into a [n, 2, 3] matrix, which can then be used with `kornia.geometry.transform.warp_affine` along with the corresponding images to perform batch alignment tasks.