Forum

Author Topic: Choosing optimum dense cloud quality  (Read 5286 times)

DCK

  • Newbie
  • *
  • Posts: 46
    • View Profile
Choosing optimum dense cloud quality
« on: September 22, 2014, 07:48:24 PM »
I have been surprised to discover that "High" and "Ultra High" dense cloud quality settings do not necessarily produce the most accurate geometric reconstructions. An example is provided in the attachment. While neither low nor high quality is perfect (it's a difficult surface, and I need to light better I think), Low is clearly a closer representation of the surface. Both models have the same polygon count.

For other objects, I have found High superior to Low.

My question is whether there is a way to know in advance which cloud quality setting will be optimal, or is this trial and error for each object (or perhaps each class of objects)?

Other ideas for getting a smoother 'vault' are very welcome.

Thanks.
« Last Edit: September 22, 2014, 07:50:14 PM by DCK »

bigben

  • Sr. Member
  • ****
  • Posts: 406
    • View Profile
Re: Choosing optimum dense cloud quality
« Reply #1 on: September 23, 2014, 07:51:58 AM »
The settings are purely point density, so you may find that the increase in resolution also shows up more of the noise in your data. This in turn may influence the surface reconstruction of the mesh. 

If you want a bit more control over the mesh from the point cloud you could export the points and do the reconstruction in Meshlab, where you can specify the individual settings for a surface reconstruction.  I've done this for some objects and it works well, although Meshlab is not muli-threaded and doesn't like handling really big point clouds in my experience. You can then import the mesh and continue the workflow.

For the most part though I find improving the end result usually means improving the input images. Shooting RAW, TIFF vs JPEG, effects of sharpening, lighting etc... If you're already at the top end of all of that, then I'd look at increasing the magnification of the images (get in closer) and lots of overlap. With experience you get a feel for what level of detail you'll get from different camera setups and shooting strategies, but there are a lot of variables so it's a little hard to give a quick answer.

Patribus

  • Full Member
  • ***
  • Posts: 170
    • View Profile
Re: Choosing optimum dense cloud quality
« Reply #2 on: September 24, 2014, 11:53:11 AM »
I tried this but I'm having a hard time to generate a mesh in meshlab which is near to as good as in PS.

I work with trees as complex 3D models and I do not know how to create a good mesh in MeshLab.

Does someone have experience with creating complex meshes?


Also, if I was only to import  a edited point cloud in meshlab into PS, this is not supported.

Cheers

bigben

  • Sr. Member
  • ****
  • Posts: 406
    • View Profile
Re: Choosing optimum dense cloud quality
« Reply #3 on: September 25, 2014, 12:47:33 AM »
Importing point clouds into PS isn't an option, but you can import a mesh.
I use a Poisson surface reconstruction for most of my meshing in Meshlab.

These are my typical settings...
Octree Depth: 12
Solver Divide: 9
Surface offsetting: 1
You can use higher numbers for the first 2 but it increases processing time.

From there it comes down to Samples per node.  The more samples per node, the smoother the mesh but the lower the resolution. What number to use depends on the density and complexity of your point cloud. In your case, it's going to be the density of the point cloud where your most complex shapes are. If there aren't a lot of points defining your trees then you'll have to use a low number.

Having generated the surface you need to remove the larger polygons that are generated to create a solid object.
Filters > Selection > "Select faces with edges longer than..."
Delete both faces and vertices
(It would be nice if PS had this so you could clean up the edges of arbitrary meshes.

After that I  resample the mesh to make the polygon density more even.
Quadratic Edge Collapse Decimation,
Quality Threshold: 1
Preserve Normal, Preserve Topology, Optimal position of simplified vertices, Planar simplification, Post Simplification Cleaning
I usually specify a target number of faces or a percentage

Then export the mesh and import int PS.
« Last Edit: September 25, 2014, 12:52:06 AM by bigben »