Cycles Development Updates

General Forums Latest News
2305

I’d love to, but I don’t have the time next to university :frowning:

15 Likes
2306

Oh, I’m sorry that you are unable to apply for that vacancy due to lack of time.
But I understand, the university is a good reason and should be a priority. Good luck to you.

3 Likes
2307

Please, some good news for cycles finally plz…
How is Stefan Werner doing? :grin: Maybe joining blender devs?

2308

Stefan works at Tangent Animations, mainly on Cycles.

4 Likes
2309

Open Image Denoise hit 1.3.0. Many improvements were made, i’m particularly interested about “Improved sharpness of fine details / less blurriness”, i often find OIDN overblurring cloth/textile textures, hope it’s now solved!

Changes in v1.3.0:

  • Improved denoising quality
    • Improved sharpness of fine details / less blurriness
    • Fewer noisy artifacts
  • Slightly improved performance and lowered memory consumption
  • Added directional (e.g. spherical harmonics) lightmap denoising to the RTLightmap filter
  • Added inputScale filter parameter which generalizes the existing (and thus now deprecated) hdrScale parameter for non-HDR images
  • Added native support for Apple Silicon and the BNNS library on macOS (currently requires rebuilding from source)
  • Added OIDN_NEURAL_RUNTIME CMake option for setting the neural network runtime library
  • Reduced the size of the library binary
  • Fixed compile error on some older macOS versions
  • Upgraded release builds to use oneTBB 2021.1.1
  • Removed tbbmalloc dependency
  • Appended the library version to the name of the directory containing the installed CMake files
  • Training:
    • Faster training performance
    • Added mixed precision training (enabled by default)
    • Added efficient data-parallel training on multiple GPUs
    • Enabled preprocessing datasets multiple times with possibly different options
    • Minor bugfixes
23 Likes
2310

That’s not places I’m too worried about noise. I use cryptomatte to do some selective/reduced denoising.

2311

Hi is it possible to update this in 2.91? or is in automatic in latest experimental build ?

1 Like
2312

Hi, the developer update the compile script to OpenImage Denoise 1.3.
It seams the prebuild library is not updated at moment.

https://developer.blender.org/rBf5c781af5643b4660c0ce27468221b37248d8372

Cheers, mib

5 Likes
2313

OIDN is ready for master: https://developer.blender.org/D10406

Cheers, mib

10 Likes
2314

It would be great if this could be added to 2.91 upwards.

2315

Hi, I guess this will not happen because they update a lot of other libraries too, including OSD, OpenImageIO, and a major update of Python to 3.9.
This may break older addons and other things, I fear.

Cheers, mib

1 Like
2316

Are there informations about the changes, and how this new version will be better in Blender?

2317

Hi, iirc there was a post about in https://devtalk.blender.org but I cant find it at moment.
Some information on the Luxrender forum:

https://forums.luxcorerender.org/viewtopic.php?f=5&t=3153

Cheers, mib

1 Like
2318

Dwivedi guiding for path-traced subsurface scattering has just been committed!

Quote from Lukas Stockner:
“In my testing with the Rain character model and a simple lighting setup, the path-traced SSS is now actually less noisy than the Christensen-Burley approximation at same render time while of course still being significantly more realistic.”

https://developer.blender.org/rB2f6d62bf88621c9a33aa01fae6548d89d0eaccd9

8 Likes
2319

Sorry for the delay on this, I completely forgot to commit it :sweat_smile:

14 Likes
2320

For those of us who value realism, it is good that surfaces with Random Walk SSS will no longer be a bit noisier than others from an objective standpoint.

Having especially noisy surfaces due to a lack of efficient sampling can really slow down a render even with adaptive sampling (as the threshold needs to be a bit lower as a result). Now to see if Lukas can also pull off sampling improvements for multiscatter GGX without sacrificing much of the physical correctness.

2 Likes
2321

Hi, a small information for Windows 7/8 user:

Announcements

  • Python 3.9 support has landed for 2.93 alpha. This has ended Blender support for Windows 7 and 8. The new minimum requirement is Windows 8.1.

Source: https://devtalk.blender.org/t/15-february-2021-upcoming/17475

Cheers, mib

3 Likes
2322

Well, physical correctness isn’t really a gradual thing - either it’s correct or it’s not. The technique that I’m looking into is an approximation, so it’s not really correct.

Then again, the exact directional distribution depends on the exact microfacet geometry. When I say that the current model is physically correct, I mean that it gives the exact result assuming a GGX distribution, but that itself is just a model.

So, you could argue that the approximation that I’m going for is just a physically correct solution for a type of surface roughness that’s not quite GGX.

In the end, what’s important is energy conservation, and it does provide that. In practice, I doubt that anyone will notice the slight difference in directional distribution.

6 Likes
2323

I think our expectations of physical correctness is indeed a gradual thing. Remember when “juggler” was considered physically accurate because it showed raytraced reflections? Followed by actual area lights, radiosity, caustics and so on. We’re still not doing “anisotropic diffuse” (think fabric rotated 180 degrees against the nap direction), spectral shading, microgrooves/glints, correct anisotropic glossy (also has limitations), complex fresnel, light polarization, more advanced (multi) lobes, bidirectional tracing for improved caustics patterns - a lot of the modern stuff. Rendering is all about approximations (we’re still doing diffuse+glossy with limited volumetric interactions), it’s never going to be physically accurate. I would even say microfacet theory is limited as an idea (it’s not just me) as it will never capture specular loss due to internal micro/nano structure, but instead only account for a modified normal. Don’t get me wrong, Cycles is doing perfectly fine and then some for what I personally need it to do.

2324

Glints would be something Cycles would definitely benefit from - it has lots of real world applications from glitter and car paint to frost and snow, brushed metals and ocean surfaces.

Lots of papers on the subject here:

https://sites.cs.ucsb.edu/~lingqi/

2 Likes