Should I build dence point cloud at all now?

[kirk]

I see I can build a mesh from depth maps .  So for a best quality possible is it necessary to build dense cloud ?   
 Would it provide extra details if I do the mesh using depth maps anyway?
   Maybe I just need to do the depth maps in higher quality and that's all?

[cbnewham]

I dropped using dense clouds long ago. I found the depth maps to be far more accurate and the whole process is quicker.

I find it surprising so many on here still use dense clouds - I imagined this was due somehow to doing drone-based photogrammetry.

[kirk]

I imagined this was due somehow to doing drone-based photogrammetry.

Wonder how it's related?  Not good enough photos from drones?   

[cbnewham]

I imagined this was due somehow to doing drone-based photogrammetry.

Wonder how it's related?  Not good enough photos from drones?   

I was thinking that maybe with all the soft edges caused by dense foliage, grass, etc, that maybe a dense cloud was better. I don't do drone photography and my photogrammetry is restricted to sculpture, so I don't really know what issues drone photogrammetry faces. From a cursory glance at the postings I've seen that most references to dense clouds come from drone photographers.

Perhaps some of those using dense clouds can enlighten us as to why they don't use the depth map capability?

[Paulo]

Hi all,

when using drone photogrammetry for mapping purpose (i.e. basically nadir looking imagery), the main products are DEMs and OrthoMosaic. And the basic and most precise input for DEMs (surface or terrain) are dense clouds and not meshes. Especially if you need to generate a Terrain model where dense cloud classification is an absolute prerequisite.

So for these applications, the idea of using just depth maps based  mesh generation does not apply....

Maybe this makes more sense....

[kirk]

Thanks Paul.   for the explanation

[kirk]

    Couple years ago I did a 400 mil mesh  comparison  Reality Capture vs then Photoscan   with dense cloud/height field method  and got an impression Photoscan makes a bit of more hi-res mesh.  I recall I saw a tiny veins in fallen leaves around rocks from my foveon matrix camera and Reality Capture had filtered them out or something.   

Now I did another test  on another subject with depth maps only and it looks a bit other way around.  Might be just an illusion or some filtering settings.  It's very close  actually.

Still I wonder If dense cloud /Height-field   could provide better result ?  It  takes more time obviously, sometimes forever,    but I wonder what method would  allow to squeeze some extra  crispness? 

[wojtek]

In 1.7 the depth maps catched up to the dense cloud resolution (dense clouds used to be considerably sharper than depth based results).

[kirk]

In 1.7 the depth maps catched up to the dense cloud resolution (dense clouds used to be considerably sharper than depth based results).

My last test was in current 1.7.1 standard version.   Is there any special tweak to increase sharpness?       Again,  am not absolutely sure , after all it's pretty close to Reality capture, maybe just  random fluctuation.

[JMR]

For me the main reason to use dense cloud rather than depth maps is the ability to apply classification and also because it is easier to remove unwanted parts from a dense cloud than editing a mesh. Better mesh editing tools and improved selection tools would be greatly welcome.

Geobit

[dpitman]

So good to see @Paulo & @JMR's comments. For use cases where classification is needed (drone photogrammetry in my case) I think the dense point cloud workflow is the only option.

[c-r-o-n-o-s]

I found the depth maps to be far more accurate and the whole process is quicker.

That's what I always thought until now. But even if I reuse the depth images, DEM creation with the point cloud is much faster than from the depth images.