What is work-flow for super hi res scans?

Follow

kir238

• July 27, 2017 10:23

I mean RC is really capable for billion of poly export. ( I did 45 GB mesh and have no idea what to do next)

So what is optimum way to deal with it? What soft is better for displacement baking, the one that could bake UDIM depth textures from hi res mesh ? Vertex color vs texture export? Parts vs the whole thing ?

0

• 

  ssh4

  • July 28, 2017 03:10

  Hello!
  
  It hard to tell without see what kind of object you have.
  
  But in most situations 200-500Mln poly mesh from High reconstruction can be decimated (simplified) 5-20 times to something more usable (10-50mln poly) without lost details.

  0

  Comment actions Permalink
• 

  kir238

  • July 29, 2017 21:39

  Are you sure? I actually think the crispness of of small surface details: pores, cracks. paint flakes etc, is getting degraded after simplification more than 20% . not times. They rather dissolve among random surface noise. Am I wrong?
  
  Much the same as in Zbrush where decimation works good only on hard surface man made objects with a lot of flat simple surfaces .

  0

  Comment actions Permalink
• 

  herveMaxwell

  • July 30, 2017 08:57

  we are supposed to get Displacement baking directly in RC... 8-)
  
  at least I recall it was a plan... no idea if it still is.. nor when to expect it..

  0

  Comment actions Permalink
• 

  kir238

  • July 30, 2017 15:59

  That's would be a revolutionary , greatest thing ever :) I spend more time baking textures than for photogrametry itself.

  0

  Comment actions Permalink
• 

  ssh4

  • July 31, 2017 03:45

  kir238 wrote:

  Are you sure? I actually think the crispness of of small surface details: pores, cracks. paint flakes etc, is getting degraded after simplification more than 20% . not times. They rather dissolve among random surface noise. Am I wrong?
  
  Much the same as in Zbrush where decimation works good only on hard surface man made objects with a lot of flat simple surfaces .

  
  
  16K texture have 268Mln pixels. This mean if you can store tangent space normals for 268Mln poly for 16K. But this will be same as a noise if all pixels will have different values. So in reality good normal map must have data from less poly on 16K for looks nice on 3D renders. Probalby 3-9 times less. So for 16K normal maps - you can store information only for 30Mln poly.
  
  Also... raw billions poly mesh have too much noise, that will redused first, details will be remain after decimation.

  0

  Comment actions Permalink
• 

  kir238

  • July 31, 2017 10:42

  Did you do a comparative tests with RC? I did years ago with Zbrush decimation and had an impression that decimation tends to mess those tiny details unless you find a very precise limit.
  
  I bet it all depends on exact settings and how decimation filters those small imperfections but I am not sure I understand your pixel math logic. Imo for nice looking normal map each pixel normal should be rather interpolated from 4 polygons for each pixel.
  At least in my experience more mesh resolution allows to distinguish so subtle depth differences as paint flakes on a wooden boards. Otherwise I see just random noise. And this noise is much easier to deal with in Photoshop on displacement texture since you can adjust it interactively.
  
  Maybe I just deceive myself I am not 100% sure

  0

  Comment actions Permalink

Please sign in to leave a comment.