memory and masked objects

[Kjellis85]

Does masking objects reduce memory consumption in mesh generation? According to the http://www.agisoft.com/pdf/tips_and_tricks/PhotoScan_Memory_Requirements.pdf the estimated consumption is based on 12 mpx images, but it does not state in what format they are (tiff, jpg). If masking does affect memory consumption, is it possible to calculate the amount needed?

[sc]

Mesh generation is based only on your dense cloud as far as I know, masking is only used during alignment and (?) dense cloud generation

[Kjellis85]

If I understand correctly, the mask only limits the pointcloud(s), not the mesh, in a direct way. But is it possible to estimate the memory consumption in mesh generation based on the "reduced" pointcloud that is created from masked images?

If I have X number of Y mPx images and i mask aproximately 50% of the image, I will then get a pointcloud with 1/2 the number of points compared to one where the entire image is used*. This can be used to calculate memory consumption based on the memory table by multiplying images with mPx not coverd in mask, ie X(1/2Y). So where I to use 12mpx sensor and mask 50% I could assume that I could process twice the amount of images compared to the table at the same memory cost?

*I understand that not all pixels in an image will create a point in the pointcloud...

[bigben]

It's based on the number of points.  Masking will reduce the number of points, but quantifying how much that reduction is, or how many points will be generated in the first instance depends on many other factors.

File format of images only affects how much disk space they occupy. Memory consumption is based on pixel count and bit depth making file format irrelevant in that case.

[Kjellis85]

Hmm, that is pretty much what I thought. However, if the number of points is in the dense cloud is a determining factor for memory consumption during meshing, is it not possible to create a plugin or feature in photoscan, or a calculator of sorts, that can calculate a rough estimate of memory consumption at the different qualities (high, ultra high etc.).

[Alexey Pasumansky]

Hello Eric,

Memory consumption cannot be estimated just based on the number of points in the dense cloud, as the surface maybe plain and regular, or elongated, or very complex 3D shape. So it wouldn't be possible to get the approximate amount of required memory before the processing is started.