Optional field-of-view correction

[gdevenyi]

In some cases we may have images which don't cover the full FOV of the average.

Assuming the masking is handled properly (so that edges don't get unduly deformed), we can weight the average by the FOV of each input image so that areas with less data don't get unduly downweighted.

[niksirbi]

This would be very nice. It's common for posterior parts of the cerebellum (or anterior parts of the olfactory bulb) to be missing in some of the images we work with, i.e. the FOV varies a bit across samples.

[gdevenyi]

Good to know I have a tester available because I don't have data yet that fits this.

This is inspired by

1. Build template normalize weights ANTsX/ANTsPy#730 (a "mostly declined" improvement to the ANTs base pipeline)
2. A critical evaluation of how the CCFv3 was constructed (completely separate project addressing some of the deficiencies coming...eventually)

@niksirbi When I actually have the bandwidth to hack on this would you be willing to share an example dataset?

[niksirbi]

Assuming that time will be in the future, me and @alessandrofelder should be able to put something together.
We are currently using modelbuild.sh to construct several templates based on serial-section 2-photon images, and any improvement that makes the pipeline do something graceful about FOV variations would be beneficial for us.

Incidentally, the preprint on the first template we made with this approach came out today: https://www.biorxiv.org/content/10.1101/2025.03.04.641293v1