Author Topic: Reinhard Luminance vs Reinhard RGB color mapping + FStorm discussion  (Read 8302 times)

2016-10-05, 18:00:50

JakubCech

  • Active Users
  • **
  • Posts: 126
  • jakubcech.net
    • View Profile
    • jakubcech
Lately I have been looking closer at Modo´s awesome meshfusion and stumbled upon rendering settings - where two Reinhard modes are present - Reinhard luminance (precisely according to paper) and Reinhard RGB. For both only burn value is available. After comparing I think Reinhard RGB is doing overally a lot more pleasant color distribution - color casts look much more natural IMO (picture 1 attached). Autodesk Vred, which is pretty awesome as well, has these colormappers, I rendered a car in EXR and made comparison in colormapping Vred(Luminance Reinhard), Vred(RGB Reinhard), Corona Highligh compression (with DeadClowns Fusion script https://corona-renderer.com/forum/index.php/topic,1807.0.html) and Arion FX Reinhard- (picture 2 attached). Arion FX Reinhard is the same as Reinhard Luminance. The most prominant difference is in blue cast on car (nicely white in Reinhard RGB) and on the ground color. In interior scenario its clearly visible how better RGB is. I attached rar with pictures for better comparison. Overally I think Reinhard RGB does better job in color cast color mapping and overall color distribution/light perception though have not found anymore about what it actually is. Corona has the best shadow / highlights distribution IMO. Maybe it RGB thing can get mapping to better level.

EDIT: Pictures of Corona mapping here are wrong - I did not use linear space in Fusion. Corona Color mapping is identical to Reinhard RGB.
« Last Edit: 2016-10-11, 08:36:58 by JakubCech »

2016-10-05, 21:18:09
Reply #1

Dionysios.TS

  • Active Users
  • **
  • Posts: 552
    • View Profile
    • Personal Portfolio
Interesting stuff, thanks for testing!
Responsable d'Imagerie
Renzo Piano Building Workshop / Paris

https://dionysios.myportfolio.com/

2016-10-05, 21:23:49
Reply #2

romullus

  • Global Moderator
  • Active Users
  • ****
  • Posts: 6400
  • Let's move this topic, shall we?
    • View Profile
    • My Models
In picture22.jpg linear raw looks best, others looks pretty bad to my untrained eye. Way too much hihghlight clamping.
I'm not Corona Team member. Everything i say, is my personal opinion only.
My Models | My Videos | My Pictures

2016-10-05, 22:43:25
Reply #3

JakubCech

  • Active Users
  • **
  • Posts: 126
  • jakubcech.net
    • View Profile
    • jakubcech
Dionysios - glad to do so.
romullus - well images are done mainly to show color propagation. Precise burn value is just "something". In other words, here I am not pointing out dynamic contrast but color (as very well visible on picture1).
Jakub
« Last Edit: 2016-10-05, 22:58:39 by JakubCech »

2016-10-06, 00:31:28
Reply #4

Dionysios.TS

  • Active Users
  • **
  • Posts: 552
    • View Profile
    • Personal Portfolio
I am still having difficulty to use the filmic highlights btw.

Now we have LUT in Corona and they seem to deal better with the original HC + Contrast than with FH + FS. It's a bit weird as at the start, when the filmic highlights appear in Corona, they seem a bit better to me but now the result gives me the feeling of flattening the image at the end so I got back to the original HC again. So much confusion on this issue... Still... :(
Responsable d'Imagerie
Renzo Piano Building Workshop / Paris

https://dionysios.myportfolio.com/

2016-10-06, 01:50:28
Reply #5

dubcat

  • Active Users
  • **
  • Posts: 423
  • ฅ^•ﻌ•^ฅ meow
    • View Profile
This is not a fStorm vs Corona post, fStorms tone mapper just look and behave better in my opinion.

I've been mimicking fStorms tone mapper with CameraRAW the past few weeks.
You can't over flatten your image with this tone mapper, and I think this is one of the main reasons people say fStorm feels like a camera.
If it looks like someone dropped a nuke in your scene when you are using 0.01, something is wrong with your light / exposure settings.

The Recipe
* I use HC in Corona when I work. But when it's time to save, I switch to 1 HC (Linear)
* Open the 32bit .hdr/.exr in CameraRAW.
* Change Process to 2010, set Recovery to 35 and Contrast to -4.

It wont match fStorm 100%, but it's much closer than Corona HC. Download the .xmp and give it a try on your car render!



EDIT:
I have attached the .xmp settings file. Because CameraRAW has some weird default settings that need to be changed.
Just load the .xmp file here and everything will be good. 35 Recovery is max, so you can use anything from 0 - 35 Recovery.



1.75 HC in Corona is the closest we can get to max "0.01" in fStorm.  But Corona boost the darks and don't have that sharp knee at the end of the curve.

1.75 HC Corona



0.01 fStorm



Comparison (Yellow is Corona, Blue is fStorm)

« Last Edit: 2016-10-06, 05:03:20 by dubcat »
             ___
    _] [__|OO|
   (____|___|     https://www.twitch.tv/dubca7 / https://soundcloud.com/dubca7 ( ͡° ͜ʖ ͡°)

2016-10-06, 11:55:27
Reply #6

JakubCech

  • Active Users
  • **
  • Posts: 126
  • jakubcech.net
    • View Profile
    • jakubcech
Well this was not meant to be FStorm thread but I made my tests anyway as I installed FStorm few days ago so we can discuss as well.
My total conclusion would be that overall "look" of FStorm comes from different color behavior before color mapping (from calculations) and different materials (BDFR) behavior (for example Roughness). Mapping is NOT as much contributing to the difference.
All tests were done with all effects (contrast, glare. flare, vignette) turned off.
Comparison1.jpg - FStorm default materials Roughness is 0,8. FStorm and Corona materials look very similar with FStorm Roughnes 1.0, going lower with Roughness makes material fade faster. - one of many first glance differences as default Roughness value is 0,8.
Comparison2 - Corona 3000K ligh is way different in terms of color to FStorm 3000K light. also overall materials/BDFR/Color calculations behavior is way different.
Comparison3 - Sphere with precise emmision value HSV 16.216.254 - FStorm has different color calculations and overall light distribution / BDFR.
In Comparison 2 and 3, very well vosoble color differences are on the second (blue) cube.
All comparisons are different even in Linear RAW (no color mapping) = these differences which are pretty huge are not connected to tone mapping but rather to calculations / BDFR model / light calculations. I have not done more tests but I can expect many other differences.
I recommend looking at pictures in attached rar where you can switch between each and see the differences better.

As I have been using custom LUTs made in AE plugged into VFB+ (waiting for Corona 1.5 LUT support, thanks god) for almost 3 years, I got this idea how to copy color changes of tone mapping from any renderer. I rendered a 32bit pattern in LUT Buddy in AE (600x600px), then resized it in PS to 7200x7200 with Nearest Neighbor (not Bilinear! - this smoothes pixel what we dont want) and projected it on plane in 3Ds Max. Resizing is needed as 600x600px pattern is usually very blurred in any renderer. Then made this plane Emmision this pattern at value 1,0 and made ortho camera looking at it - and render in high resolution (4800x4800 in my case). Then again - put this pattern to PS, downsize to 600x600 (again Nearest Neighbor). Then if needed apply gamma (I had to) and finally plug to AE and generate LUT with LUT buddy. This was not needed in this case but may be helpful idea in some cases :) (attached pictures are squized to 8bit jpeg). I did not use Screen environment as FStorm does not support Bitmap node.

Jakub
« Last Edit: 2016-10-10, 13:00:00 by JakubCech »

2016-10-06, 21:07:15
Reply #7

dubcat

  • Active Users
  • **
  • Posts: 423
  • ฅ^•ﻌ•^ฅ meow
    • View Profile
We are talking about two different things.

I'm talking about how fstorms tone mapper behaves more (in my option) like a tone mapper in a real camera (emulating camera behavior).
You can compare these tone mappers on a real world photo.
Take a picture in RAW + JPG and run this script on your RAW
Code: [Select]
"dcraw-9.27-ms-64-bit.exe -v -w -H 0 -o 1 -q 3 -4 -T *.NEF/.CR2"Load that TIFF into Corona Frame Buffer and fStorm Frame Buffer, now you can play around with the tone mapper and compare them to the original jpg (camera tone mapper).

You are talking about light and materials (emulating real world).
             ___
    _] [__|OO|
   (____|___|     https://www.twitch.tv/dubca7 / https://soundcloud.com/dubca7 ( ͡° ͜ʖ ͡°)

2016-10-07, 00:52:00
Reply #8

Captain Obvious

  • Active Users
  • **
  • Posts: 168
    • View Profile
The Reinhard RGB tone mapper and the highlight compression in Corona are pretty much identical. They will produce identical results, but for different parameter values.

2016-10-10, 12:10:29
Reply #9

JakubCech

  • Active Users
  • **
  • Posts: 126
  • jakubcech.net
    • View Profile
    • jakubcech
Dubcat can you please clarify how to load tiff into VFBs?
Captain can you please elaborate what you mean by different parameters? - if you observe it closely (from attached rar) you can see Reinhard RGB is painting car white (nicely adjusted color cast IMO) whereas Luminance Reinhard + Corona leave it somehow blueish.

2016-10-10, 12:27:46
Reply #10

dubcat

  • Active Users
  • **
  • Posts: 423
  • ฅ^•ﻌ•^ฅ meow
    • View Profile
#01 Change Environ to Screen and disable Filtering



#02 Change Image Filter to None in System



#03 Set Pass limit to 1, this will give you pixel perfect result inside VFB.



#04 Throw the Bitmap into Direct Visibility Override



#05 Remember to change Output Size to match the bitmap



06# Now crank that Bloom intensity up to at least 2000, we want people to think there is an alien spaceship outside the window.

             ___
    _] [__|OO|
   (____|___|     https://www.twitch.tv/dubca7 / https://soundcloud.com/dubca7 ( ͡° ͜ʖ ͡°)

2016-10-10, 13:38:30
Reply #11

peterguthrie

  • Active Users
  • **
  • Posts: 239
    • View Profile
    • Peter Guthrie Visualisation
tried this as I was curious but the bloom and glare dont work.. should they work?

EDIT: i need to stopping posting message so quickly - bloom does work, it was just slow as my image size was 8k

2016-10-10, 13:40:34
Reply #12

Captain Obvious

  • Active Users
  • **
  • Posts: 168
    • View Profile
Captain can you please elaborate what you mean by different parameters? - if you observe it closely (from attached rar) you can see Reinhard RGB is painting car white (nicely adjusted color cast IMO) whereas Luminance Reinhard + Corona leave it somehow blueish.
MODO's Reinhard RGB produces identical results to Corona's Highlight Compression, but use a different input parameter to do so.

A tonemapping amount of 25% is identical to a highlight compression of about 1.156. 50% is 1.414, 75% is 2 exactly. 95% is about 4.472.

The only difference is that a tonemapping amount of 100% is an highlight compression amount of infinity.

2016-10-10, 13:55:49
Reply #13

JakubCech

  • Active Users
  • **
  • Posts: 126
  • jakubcech.net
    • View Profile
    • jakubcech
Okay guys - Thanks Dubcat for tutoring, I managed to load EXR from first post (from Vred) and apply HC in Corona and yes - its identical to Reinhard RGB (thanks Captain Obvious). I guess Fusion script is implemented differently as I obtained different mapping? Dont know - I may have done some mistake. Its sorted out now anyways.

2016-10-10, 19:02:08
Reply #14

DeadClown

  • Global Moderator
  • Active Users
  • ****
  • Posts: 1445
    • View Profile
    • racoon-artworks
I guess Fusion script is implemented differently as I obtained different mapping? Dont know - I may have done some mistake.
Since I got the info directly from Ondra it should be the exact same thing :) . Just make sure you're using the macro in linear color space, not on a gamma corrected image.
Any sufficiently advanced bug is indistinguishable from a feature.