Search code examples
3drendermeshraytracing

Fast way to render a small preview of a complex 3D model.

I am looking for a very fast way to render a small preview of a complex 3D modell.

  • The Modell comes as stl or ply.
  • A modell will have between 300.000 to 500.000 faces.
  • The rendering process will run on a headless linux server without a GPU.
  • The rendering process can use only one core/thread.
  • Required time should be under 15 seconds.

The resulting image will be small, about 300x300 px. It does not need to have shadows or other pseudo realistic features.

What I tried so far:

stl2pov.py (stltools) -> povray: Result looks good and rendering time is below 15s, but Povray needs a minute parsing the pov file created by stl2pov. If I use mesh2 format parsing is much faster, but the result is gibberish. I believe this is caused by a bug in stl2pov. Is there another stl/ply to pov converter which produces mesh2?

Blender with cycle renderer: It all depends on the number of samples. A sample number of 150 gives an quite noisy but ok result, but still needs 30 seconds. But as I am a Blender newbie and this beast is quite complex, maybe there are some tricks I did overlook.

It seems that raytracing solutions are too slow and are an overkill anyway. All other solutions I found depend on a GPU for speed.

Do you have any ideas how to solve this problem.

Solution

  • For a size of rendering like 300x300 there's not much sense using 300k+ triangles so you may try to simplify your mesh to much lower triangle count without noticeable visual difference and try rendering this mesh with the solutions you've already tried. This should lower your rendering times noticeably.

    You may also try to just ignore triangles that are smaller than some value and see if that works for you and your models.

    If you need to write entire rendering code and want to make it as simple as possible (or get some working code early) you may try to draw just a mesh vertices as single pixels. With such dense model it may work quite well and will be simple to implement, even with Z-buffer.