Also available for macOS now: GraphicAll
3 Likes
Thanks for your support by making these builds available on MacOS!
4 Likes
A new build is available on GraphicAll. What’s new:
- Branched Path Tracing is now working
- Updated OCIO config to be more compatible with already existing files. Colorspaces were renamed to match those from the
masterbranch to avoid wrong colorspaces being assigned to the textures. Render results should now be much closer to the original ones out of the box. - Fixed crash during light initialization
- Updated Blender
15 Likes
Hi, I follow with great interest this topic even if parts of the technical stuff are beyond my knowledge and I want to share this Two Minute Papers presentation about new(2020) research on light transport simulation that could be at least inspiring for this project.
1 Like
That’s quite an unrelated change to Spectral rendering although yes, I’d also love to see something like this in Cycles. The question is whether it’s compatible with Cycles’ light-path- and node-based architecture, which apparently was the biggest issue with stuff like this in the past.
1 Like
I’ve just downloaded the 7zip and it extracted and launched without any problems. Try re-downloading it, if that doesn’t help I’ll reupload the file.
Just tried re-downloading in case that was it a second ago and it happened again :-/
I’ve repacked and uploaded the build, this should hopefully work now.
Thanks it works now!
3 Likes
I think you mean: https://www.youtube.com/watch?v=Popg7ej4AUU which displays results by Zeltner et al (students of Wenzel Jakob). You can find the (prepub) paper “Specular Manifold Sampling for Rendering High-Frequency Caustics and Glints” here: https://rgl.epfl.ch/people/wjakob
5 Likes
A new build is out, as usual. Changes this week:
- Changed default display device to “sRGB Display” in the OpenColorIO config. This brings render results even closer compared to the reference renders.
- Improved Principled Hair BSDF compatibility. It doesn’t look identical to what it should be, but it’s much better than it was before (completely broken).
- Fixed Toon BSDF
- Updated Blender
Barbershop scene comparison, rendered with master and spectral-cycles builds:
18 Likes
No one is using an sRGB display in the vast majority of cases, so the output is likely broken.
looks like the second one is slightly more bluish
The blueish looks like it is originating from skylight?
Yeah if you flick between them the largest difference is in the patch of light coming from outside. Environment background importance sampling is currently broken, which could be a cause for this.
EDIT: Actually looking again, it seems like something is up with the texture on the floor. There’s the largest difference in that area. Not sure what could explain that.
1 Like
Some parts also just look straight up darker. Notably the glass table in the foreground.
But to me it looks like every bit of red got ever so slightly muted. For instance, the leftside wall is slightly warmly lit but looks a bit cooler in the second image.
The sofa and the chairs (especially the one closest to camera) also look noticeably bluer or darker.
This might all indeed be from the skylight though.
Easy way to test is just to put a bunch more samples into each. If they get more similar it’s just a sampling issue, if the error persists there’s a change in material or light behaviour somewhere. Also keep in mind they won’t be 100% identical because one is spectral, bounce light will have a marginally different quality.
1 Like
It’s entirely possible that this is spectral and expected, yeah. Especially green is gonna be more obvious in bounces as I think my tests have shown before, because the spectral overlap of that with both red and blue is quite significant. (If you’re using the current RGB-to-spectral conversions)
It doesn’t really look like it’s a noise issue though. These aren’t necessarily regions with a lot of noise.