I have attempted a reproduction of Bruno Cabasson's "Extreme photons"
using spectral rendering. Bruno's originals are at:
 https://news.povray.org/web.4d54f6b329b411e54aa45fdf0%40news.povray.org
 https://news.povray.org/web.56b2713b1c633944f94e287a0%40news.povray.org

Image rc3-extreme_photons-pre.jpg is a 3-channel RGB preview render in
which I used POV-Ray's native dispersion to calibrate the optical
properties of the glass used in the spectral render.  Image
rc3-extreme_photons-sr.jpg is the spectral render.

Without access to Bruno's source code, I had to guess at the placement
of the objects.  Hence, there are some differences, especially at the
upper right.

But I was also limited by SpectralRender's wavelength resolution, which
is fixed at the equivalent of 36 dispersion samples.  In addition, the
deepest violet wavelengths are nearly invisible, so I stretched out the
dispersion such that the 2nd mirror only catches wavelengths of 430 nm
and longer, leaving effectively only 31 dispersion samples.  It is clear
from the images that at least twice as many dispersion samples are
needed for smooth transitions in the blues and violets.

One change I have contemplated is evenly spacing spacing out the
wavelength IORs.  IORs are non-linear with respect to wavelength, and
for most real-life materials, the violet rays are spread out more than
the red rays.  For this scene, I used an exaggerated version of BK7
glass, which exhibits this property.  Forcing a linear correspondence
between IOR and wavelength should bring the violet rays closer together,
although the RGB preview (which uses dispersion_samples 36) shows that
merely adjusting the spacing would not be enough.  Drastic editing of
SpectralRender would be needed to increase the number of dispersion samples.

One odd glitch in the images is that the media photons are not reflected
smoothly in the mirrors.  Anyone have an idea what causes that?

Light and optical properties:

  photons { spacing 0.02 media 300 autostop 0 }
  interior { ior 1.5 dispersion 1.1 } // RGB preview only
  light_source
  { -14 * y, [daylight 6504 K] * 1.5
    cylinder radius 0.01 falloff 0.4
    parallel point_at 0
    media_interaction off
  }
  white point: daylight 6504 K
  gamut mapping: Y-preserved

RGB preview statistics:

  Max Level: 50/50
  Number of photons shot:          804403
  Surface photons stored:           47938
  Media photons stored:          56110011
  Gather function called:        99876683

  Render Time:
    Photon Time:      0 hours  0 minutes 17 seconds (17.696 seconds)
                using 10 thread(s) with 22.581 CPU-seconds total
    Radiosity Time:   No radiosity
    Trace Time:       4 hours 30 minutes 37 seconds (16237.189 seconds)
                using 7 thread(s) with 102226.472 CPU-seconds total

I used only 7 threads because I wanted to keep using the computer while
it was rendering, and even with nice -n 19, renders can interfere with
some other programs.  I tried doing an overnight run, but the whole
entire computer crashed sometime during the night.

The last render block took 16 minutes to finish after all the others
were done; and based on previous runs, the last 5 render blocks probably
took around 39 minutes.

Spectral render statistics:

  Max Level: 50/50
  Number of photons shot:   804403 per animation frame, 28958508 total
  Surface photons stored:     1323 average per frame,      47623 total
  Media photons stored:    1518543 average per frame,   54667531 total
  Gather function called: 18813873 average per frame,  677299421 total

  Render Times:
    Photon Times: 0.456 seconds average per frame, 16.418 seconds total
      using 11 threads
        0.691 CPU-seconds average per frame, 24.860 CPU-seconds total
    Radiosity Times:   No radiosity
    Trace Times: 14 min 38 sec average per frame, 8 hrs 46.6 min total
      using 8 threads
        6345 CPU-seconds average per frame, 228413 CPU-seconds total

Each animation frame corresponds to a wavelength of light.  I ran this
one overnight on all 8 threads because it's an animation, and I figured
if it crashed I could easily pick up where it left off.

Spectral render final assembly statistics:

  Render Time:
    Photon Time:      No photons
    Radiosity Time:   No radiosity
    Trace Time:       0 hours  1 minutes  1 seconds (61.512 seconds)
                using 8 thread(s) with 448.695 CPU-seconds total

Post a reply to this message

Attachments:
Download 'rc3-extreme_photons-pre.jpg' (152 KB) Download 'rc3-extreme_photons-sr.jpg' (143 KB)

Preview of image 'rc3-extreme_photons-pre.jpg'

Preview of image 'rc3-extreme_photons-sr.jpg'