Under the export settings when exporting the registration and COLMAP is chosen (though I imagine it can apply to other formats), include a new setting labeled something like “Single Camera”.
The reason is that even if you group all the inputs under the same lens distortion and calibration group (which determines the camera intrinsics for all the images present - effectively assuming they’re all shot with a camera with unchanging settings), under the COLMAP format ALL the cameras for EVERY image still get their own line in cameras.txt. When the setting I’m suggesting is enabled, cameras.txt would have a SINGLE camera written.
For those curious: when running Nvidia’s 3DGUT/3DGRUT a single camera for the 750 photos in my testing and all the registration ends up at like 110MB or so of memory…with 750 cameras (even though identical) for those 750 photos it balloons to like 100GB of memory!!!
I’ve created my own script to fix this and it works well, so I thought I’d suggest it to the devs as its a very easy fix. Simply have the option greyed out if the camera intrinsics don’t all match (hovering may give the hint to group the lens and calibration groups on all images). Another way this setting can be done (in case like 2 or 3 cameras are used for all the images) could simply be: “Condense Cameras” aka combine all duplicates of intrinsics.