User:LeonardoSegovia/GSoC 2018/Reports/Week 5

Week 5

Hi everyone! This is my report for the present week.

Objectives

This week I dealt with Weeks 4-5 objectives:

• Test shader functionality. If working, release test build.
• Evaluate received feedback. Start bug fixes.

I released two builds this week:

• one with the Week 4 features
• the other almost mid-week after due to a bug (more on this later!).

Remember that I was partially out on Mon-Tue because of a public class I had to prepare.

What's been done

• In the BA thread, Zanzio found that renders were wildly inconsistent between CPU and GPU. This became noticeably if rendering with both at the same time.
  • Bug #1 was a precision issue when normalizing the coordinate frame in the presence of low roughness values (rBe2671913).
  • Bug #2, and the true reason for this, was that somehow the GPU kernel was reordering the second set of values. I don't know why it had such an effect, but it's been fixed. Thanks to nirved again for noticing! (rB2d1eef87)
• Did a bit of cleanup: store and retrieve consistently the values from the stack (rBdc2827fd).
• Merged the latest master commits (rB60a59fba, rB535e2e4a, rB9b632add.
  • Lukas changed the linear_rgb_to_gray function signature. (rBbd37b18a)

What I'll do later today, and during the weekend

Today, Brecht, Lukas, nirved and I discussed on IRC the current state of the shader. This is a summary of the proposed changes:

• Rename the Roughness parameters to (Longitudinal) "Roughness", "Radial Roughness" and "Undercoat Roughness"
• Remove the "Physical" parameterization
• Rename and reorder the rest to "Color" ("Direct coloring") "Melanin" ("Melanin Concentration"), and "Absorption coefficient"
  • The latter remains for debugging purposes e.g. if someone wants to use the coefficients straight from the papers
• Remap Melanin concentration to represent:
  • Melanin -> quantity
  • Melanin Redness -> distribution
  • Brecht asked if Melanin could be done as a percentage of saturation (0 = white, 1 = black). I know it can be done that way (V-Ray), but there is no public, reliable mapping; for example, Tungsten's melanin values have a similar representation, but Bitterli still uses d'Eon's mapping, so the Melanin value is still an absolute concentration quantity.
• Redo the randomization to a relative tolerance
  • Formula: effective_roughness = clamp(roughness * (1.0f + (random - 0.5f) * roughness_randomization/100.0f), 0.0f, 1.0f)
  • i.e. when randomized, the roughness value should be roughness ± roughness_randomization%-
• Implement randomization by directly retrieving Hair Info -> Random unless the user sockets a different source.
  • This is easy with OSL (geom:curve_random), with SVM... not so much. I do have an idea, thanks to Principled Volume.
• Reorder the sockets as follows:
  • Color
  • Roughness
  • Azimuthal Roughness
  • Undercoat Roughness
  • IOR
  • Offset
  • Randomize Color
  • Randomize Roughness
  • Random

Next up

If no bugs or feature requests arise in the next weeks, I'll begin porting the existing hair tests to the new shader. I reiterate the previous request for .blend hairstyles for this purpose -- it'll be great if you could post them to the BA thread!

Questions

None this week!