Improving 8K textures close-up (Redshift)

   2992   5   1
User Avatar
Member
28 posts
Joined: May 2020
Offline
I have rendered this simple scene with Redshift, using textures generated by processing 16K and 8K assets from NASA and https://www.solarsystemscope.com/textures/ [www.solarsystemscope.com].
I have attached three sampled screenshots from the resulting 10000x10000 rendering, which looks terrible for some reason. The textures are in PNG format, and I have attached a sample of the flat texture I used for the lunar surface.

Is there a way to improve the definition of the texture, maybe with a filter? Does RS have any specific filter that might help?

Attachments:
Capture.JPG (303.4 KB)
Capture1.JPG (242.9 KB)
Capture2.JPG (183.3 KB)
Capture3.JPG (267.6 KB)

Fabio Basile - 3D Artist

https://www.behance.net/fabiobasile [www.behance.net]
User Avatar
Member
806 posts
Joined: Oct. 2016
Offline
Hi,

the “broken” images don't look like they are showing a problem that a filter could fix, instead, they look like a stripe-offset-error in reading/displaying the actual data. It might be coming from your UVs (could, maybe, be a floating-point issue?) or could be a bug in RS.
Have you tried narrowing down if the error appears on a simplified model (like a plain grid)? If it also appears there, have you tried exposing the issue in the Red Shift community?


Marc
---
Out of here. Being called a dick after having supported Houdini users for years is over my paygrade.
I will work for money, but NOT for "you have to provide people with free products" Indie-artists.
Good bye.
https://www.marc-albrecht.de [www.marc-albrecht.de]
User Avatar
Member
6269 posts
Joined: Sept. 2011
Online
They look like floating point arithmetic errors. Is your “moon” really big?
User Avatar
Member
28 posts
Joined: May 2020
Offline
You may be onto something, and this makes total sense. I suspect that this may very well be a RS issue, as all my textures, no matter what mesh I use, present similar problems.

I have initially created the scene to reflect roughly accurate distances between the Sun, Moon and Earth, as well as their relative size, scaled down to by quite a few decimals to make it manageable, but not so small as to have the moon down to 0.0000~e scale. Still, proportionally, the distance between the sun and earth/moon is about 1 kilometer, and i have placed an RS point light inside the Sun, and have specified only the moon and earth in the list of objects influenced by the light source. I have also a light dome, which does not affect anything, as i have turned it off, and only used it to have a spherical projection of the stars, which I have generated from NASA footage, and down-sampled it to a 2Kx4K.

All meshes are textured using UDIM tiles, however the issue manifested even with regular textures.

I have eliminated all the volumes, dome light, and rendered each mesh individually, and I also tested it with sample objects, and the issue still persists.

I will bring this issue to the attention of the RS community as well, hopefully I'll get to the bottom of it, but thanks for pointing me in a good direction, very much appreciated! :-)
Fabio Basile - 3D Artist

https://www.behance.net/fabiobasile [www.behance.net]
User Avatar
Member
28 posts
Joined: May 2020
Offline
Here's a happy update:

According to a response i got in the RS community forums, a working fix is to go into the experimental section of the Redshift ROP, under System, and check “Render In Camera Space”. I have tested it an apparently the issue is gone.
Fabio Basile - 3D Artist

https://www.behance.net/fabiobasile [www.behance.net]
User Avatar
Member
806 posts
Joined: Oct. 2016
Offline
Ah, “render in camera space” sounds like they are then first applying the camera matrix to the world space construction before projecting textures - that would get around massive floating-point issues that you would have with such huge scales (assuming you are using meters / SI units in the scene).

It *might* help to use a scene dependent unit like “diameter of the sun” instead of something like meter or kilometer. However, I fear that with the differences in sizes and distances in a solar system, ANY “fixed” system will run into limitations if you want to apply really fine detailed textures - I haven't looked at your source data, but I assume you are using meter/kilometer sized pixels there and are trying to project those onto immense areas. That's always going to be a problem, don't even think about SIMULATIONS with specifically tailored solvers for such setups :-D


Marc
---
Out of here. Being called a dick after having supported Houdini users for years is over my paygrade.
I will work for money, but NOT for "you have to provide people with free products" Indie-artists.
Good bye.
https://www.marc-albrecht.de [www.marc-albrecht.de]
  • Quick Links