Best practice for chunks

[cbnewham]

Does anyone have a document or website detailing the best practice with chunks? The manual just tells you how to create them and copy cameras to them.

In my case I have a model which I can't generate in one go because the mesh creation from depth maps runs out of memory.

In the past I've loaded all the photos in, created the sparse cloud, and then duplicated the chunk several times and in each one selected from the 3D view a subset of cameras to keep active and disabled the others (keeping some overlap of selected cameras between chunks), then created the dense cloud, then merged the chunks, exported and decimated the dense cloud, then imported and created the mesh.

I'm not sure if this is the way to do it or if there is a better way. I'm also unsure if this will work using the new mesh from depth-map generation.

Thoughts (or tell me this is not the best way!)

Thanks.

[Alexey Pasumansky]

Hello cbnewham,

We hope to implement out-of-core method for the depth maps mesh generation from the depth maps in the version 1.6.0 (hopefully in the next pre-release as well), so if "not enough memory" is the only problem that you are facing, if might be better to wait a little bit till the version is released.

In general, if you are to split the model into blocks, I could suggest to use the same camera alignment (to avoid the chunk referencing artifacts) and vary only the bounding box to define the area of mesh reconstruction (you can even build the depth maps for the complete project), after that merge back all the "tiles".

[gto234]

Alexey,
what memory requirements will have the new version with out-of-core processing?
Will 32GB be enough for arbitary 3d mesh from depth maps?

[Mak11]

gto234,

Alexey mentioned in another thread that they were aiming for 16GB.

Mak

[gto234]

That's cool..!

[cbnewham]

Thanks Alexey for the advice.

I will have a go and see how this works.