right setting for RTK photos dataset

I have a dataset of photos acquired with DJI Phantom in RTK mode. During the flight the accuracy of the position was displayed around 2 cm, except some image, around 10% of photos (but I don’t know exactly which).
I mesuared some GCPs with GPS too.
I want produce a model using the high precision position of the cameras + 3 or 4 Ground Control Points + other 4/5 Ground Test
So, my questions are:

  • what is the best setting that I have to use, to produce my model for this case?
  • how can I use the high photos acuracy, if there are some photos (unknow) that have a low accuracy?

This topic will be discussed in a support ticket created by user. 

Posting the answer here for others:

Select those images with lower precision in 1Ds view in RealityCapture? If yes, I would suggest switching their Absolute pose to Unknown (Selected inputs panel / Prior pose).

If this doesn’t work, then working with GCPs should help.
To get the best results when using GCPs make sure that you have at least 4 points at the corners of your model. If you have more than 4 points, try putting the as homogeneous inside of the model as you can. This will help prevent distortion which can appear, when GCPs are not placed right.
I always recommend, when using GCPs, to use only GCPs for georeferencing, since they are usually more accurate. To do so, select all inputs and in Selected inputs / Prios pose set Absolute pose to Unknown for all cameras.