How to passfull or inhomogeneous affine matrix to `Affine`?

[jo-mueller]

I have a use case where I extract an affine matrix from a different tool, which gives me an affine matrix of the following kind for a 3D dataset:

2.2408546792146753, 0.518379728429602, -0.04612377550830799, -210.8286161082715
-0.5159947829355463, 2.2395791802696188, 0.10153382765538543, -103.19297075390104
0.0677814834389687, -0.08855614788543548, 2.2977898333064966, 4.337448299935659
0.0, 0.0, 0.0, 1.0

Now, if I chain this in a TransformSequence with a Scale transform with three scale parameters, I get a ValueError: Cannot compose affine transformations of different dimensionality error, because the affine matrix is automatically padded to 5x5 under the hood. Maybe there's a way to check if the affine is passed in homogeneous or inhomogeneous format?

For now, I can just pull out the translation from the affine (right column) and split it into an affine + translation, but it would be nice to support this.

[clbarnes]

In the current release, we don't do any padding. The argument should be the full affine transformation matrix (augmented, i.e. shape (Di+1, Do+1)), and under the hood we just validate that the bottom row is [0, 0, ..., 0, 1]. The padding logic was in an earlier version. If someone wants to pass in a non-augmented matrix (a linear map), there's the from_linear_map constructor which optionally takes a translation as well.

Could you share the transformnd version and a code snippet to reproduce your issue?

Remember that OME-Zarr affine transformations don't store the bottom row, so if that existing [0, 0, ..., 0, 1] row is in your OME-Zarr metadata, your reader will be adding another [0, 0, ..., 0, 1] row to the bottom, which would mess with the dimensionality.