GCPs and accurate reconstruction accuracy estimate

[Jernej]

Hello

On how many pictures in the dataset must you mark the control points to get the most accurate estimate of the reconstruction accuracy of the model?

I have a model that I need to reconstruct with a mean error of less than 1.8 cm. To get the most accurate report of the error should I mark the control points in all of the images where they are visible or should I leave only those that were automatically detected with the Detect Markers function?

This question has been probably been asked a million times already but for the life of me, I cannot find a satisfactory answer.

Thanks for your help and insight.

[JacquesMDC]

Hi, in  my opinion, the most GCP you mark, the better it is :-) But 1,8cm accuracy with GCP is quite good !

[Jernej]

Thanks for the answer. I've been doing some testing and comparisons with Metashape and RealityCapture. What I've seen is that RealityCaputre usually performs much faster but lacks any reliable and fast way to get good "control" over the alignment process. It has a really poor marker detection algorithm and if you want to mark all of the GCP-s it's a slog. Also, the component workflow on larger projects seems very clumsy. In this way, Metashape, at least to me, has a vastly superior workflow compared to RealityCaputre