Hi Damir,
I assume you did not acquire an RTK/PPK image dataset. If your dataset has been acquired with a consumer drone that relies solely on a standard GPS receiver, then the absolute accuracy will lie within a few meters. Note that image geotags are usually measured with an accuracy between 5 m (X, Y) and 10 m (Z).
In projects with image geolocation and without GCPs, the absolute camera position uncertainty will be similar to the expected GPS accuracy. As all images are positioned with similar accuracy, the sigma reported in the table should be small compared to the mean (which is also your case). The absolute camera position uncertainties are also expected to be bigger than the relative ones in the table Relative position and orientation uncertainties.
In your case, I would recommend using GCPs. GCPs are definitely more accurate and reliable than image geotags recorded with standard GPS. They make the model more robust and accurately georeferenced. If you are using GCPs, it is likely that some of the cameras will be moved significantly with respect to the initial geotags. This would be expected due to the low accuracy of the image geolocation.
Hopefully this helps.
Best,
Teodora