For a project of 120 images it should be possible to work without using chunks. If there are many misalignments after the initial alignment step, my usual workflow is the following:
- look for a subgroup of images which appears to be well aligned
- reset alignment for all remaining images
- individually align the remaining images one by one
- check alignment after each image alignment (both by looking at the point cloud and modeled camera positions and by looking at the numbers in the refined camera calibration - if the refined calibration parameters - in particular focal length - are more than a few per cent off the initial parameters, I assume that that camera is not well modeled and misaligned)
- if an alignment looks bad, reset that alignment and skip that image (perhaps try to align it again at the end after all other images are well aligned)
This can take some time but in my experience can result in good models even if things looked horrible after the initial image alignment. Also, try to use "separate" under refined camera calibration, in particular if you have movement within the lens systems beween photos (e.g., focussing, optical image stabilisation) which would not be taken into account if you used "groups".