Has anyone here tried building their own .json file with colmap?

[jurek0]

Hello,
I have used the provided colmap2nerf script and it seems to be working fine buttttt....
I can't help but wonder if there is an even more effective way to use colmap to extract the camera poses and other information for your .json file.
Is there anyone here that has tried to tweak some parameters in colmap to achieve even more accurate camera poses?
Has anyone tried to write their own script? Are there even really parameters to mess with and do you think the idea in itself is stupid?

My thought process here is:

1. Camera poses are extremely important for NeRF.
2. COLMAP is said to be very finicky.
3. There could maybe be a way to extract perfect camera data by tweaking some parameters in colmap BUT
4. I am definitely not skilled enough to fully understand all of colmaps functions

I have so far read the colmap tutorial for feature extraction and it says that it is possible to "manually specify intrinsic parameters" but this refers to a lab setting I think.

If you think it doesn't really make a difference if I'm using the provided script or trying to do it myself please let me know!

Thank you for your answers :)