Bounding Box size & Photo Set & Dense Point Cloud behaviour

[scanlab.ca]

Bounding box size seems to affects number of photos used when building the point cloud.

Is this the correct behavior, because when I build a small part of an object I expect the whole photo set be searched for the corresponding area inside the bounding box. But there seems to be some optimization that discards certain photos, which defeats the purpose of having a resizable bounding box because not everything withing the bounding box ends up being build.

Sounds a little strange doesn't it?


PS Version 1.0.0 build 1736.

[Alexey Pasumansky]

Hello Ruslan,

In the latest version PhotoScan uses only points from the bounding box for depth maps estimation.

However, you can reconstruct the full dense point cloud and for mesh reconstruction use smaller working volume.

[scanlab.ca]

Hello Ruslan,

In the latest version PhotoScan uses only points from the bounding box for depth maps estimation.

However, you can reconstruct the full dense point cloud and for mesh reconstruction use smaller working volume.

I think that's the issue, it does not use all points from the BB volume.  The dense point cloud estimation changes as I expand the bounding box, making me scale the BB to a much larger volume than the reconstructed surface needs to be.

P.S.
I remember someone in another thread was mentioning that their reconstruction was erroring out and you suggested making the BB volume larger. They might've have had the same problem I'm having, where the BB volume does not encapsulate or include enough photos for reconstruction. Its just a guess though.

[scanlab.ca]

Let me add this image to show you what I mean.

It's a partial screen grab of a UltraHigh build spot, which is
1. well inside of the bounding box,
2. has no masks to prevent the area from being generated
3. builds fine when doing a full high build

and, since this cut off line is so obvious(white area is the actual surface, dark area is the opposite side, not facing the camera), I make an assumption that Photoscan discards a few photos, which are actually needed to get this area built.

I wish I could build this whole thing at ultra high, so it would reconstruct everything, but I cannot due to my 24gig RAM limit. But, even then, this makes me wonder if there are other spots which PS could potentially do a better job at building, but doesn't due to the insufficient BB size that does not account for this potential photo discardment / loss.


I think I'm starting to go in circles now.