each with each vs structure from motion

there is actually a possibility to define which features are used when connecting components.
for example if you have common part of stairs with floor reconstructions you may want to mege components using overlaps which is fastest option.
on the other hand if you are looking through the windows i’d use all image features to get the best precision