Sparse and dense point cloud

[janoz]

I would like to ask if editing of the sparse point cloud does have influence on dense point cloud generation. From purely logical point of view I d think that deleted parts of sparse point cloud are not taken into account during dense point cloud generation. But roughly from "macroscopic" observetaion it seems to me as if deleted parts of sparse point cloud  in some way "reappear" in dense point cloud. Am I right?
What is the relationship between sparse and dense point cloud in the sense of their subsequent dependancy. I have a feeling that a basic source for dense point clouds are source photos rather than sparse point cloud.
Thanks

[Marcel]

As far as I know the only influence of the Sparse Cloud on the Dense Cloud is the selection of which camera pairs it uses during depth reconstruction. So if you have a lot of inaccurate points, then Photoscan might start reconstructing with two images that do not have enough overlap. (But I have no idea what the thresholds are for camera pair selection, ie. how many points do two images need to share to be considered a pair).

I haven't tried of deleting points in a specific area will exclude the images in this area, it would be an interesting experiment.

Of course the quality of the alignment is also a huge influence on the Dense Cloud reconstruction.

[Lambo]

From my experience, the sparse cloud points do have an influence on the Dense cloud generation. I keep doing tests deleting or refining the sparse clouds with the Gradual selection and it makes a noticeable difference. I also normally delete any points that are out of the masked area that still came through and it makes those areas not appear on the dense cloud.
For example, when I refine the sparse cloud with the Gradual selection, the final result is more smooth but also lacks some details where the sparse cloud had too few points.
If I don't do the Gradual selection, the detail is rougher but still has some detail on those areas that didn't have too many points on the sparse cloud.
Leo