I’ve been using RealityCapture daily since last year and after updating to RealityScan 2.0 recently I’ve seen drop in quality of alignment and mesh model with default settings and input.
Has anybody noticed it as well. Overall it works well but it looks like step backward in terms of quality.
Hello Kamil,
are you using the same settings in both version?
There is no difference in meshing process, but the alignment was changed. Are you getting similar alignment results? Can you try the old setting in RealityScan 2.0 for alignment (set Feature detection quality to Normal and align data again and then process the model).
I used default the same settings for alignment for both versions - confirmed for medium and high feature detection quality. It looks like the new algorithm is not as bulletproof when patchmatching in variable overlaps. Everything is fine when overlap is close to constant
Is it possible to share that dataset with us?