Filmgrain: fix native filmgrain by making it work in approximate PQ space and apply the same fixes to the improved one by EndlesslyFlowering · Pull Request #21 · EndlesslyFlowering/Starfield-Luma

EndlesslyFlowering · 2023-10-18T04:06:29Z

I rewrote both the native filmgrain and @clshortfuse's filmgrain to work in "perceptual space".
I use a approximation of PQ for HDR and leave it in gamma space in SDR.
Filmgrain produces better results in "perceptual space" rather than in linear.

Just realised that for SDR it's not a given that the input is in gamma.
@Filoppi are we going to do SDR in linear until output? or is that still up in the air?

…to constrain the output within 0-1)

(with some minor changes)

-Inverse tonemap by luminance (to maintain SDR hue) -Inverse tonemap highlights only (to maintain SDR crushed shadow filmic look) -Automatically find paper white multiplier by checking how much mid gray shifted with the inverse tonemapper -Improve LUTs shadow invalid colors

…Ts to have HDR/scRGB colors (beyond 0-1)

… and inverse tone mapping

…d ".editorconfig"

…st of the image will have the same look as SDR -Improve LUTs shadow gradient -Remove AutoHDR (it was just a fun test) -Clean up code

…the LUT already was (HDR only) (they are cached on the side when computing the SDR image) -Fixed Inverse Hable being unable to restore highlights, by simply using the original color instead, given that there's no post process anymore happening between tm and inv tm -Improved "INVERT_TONEMAP_HIGHLIGHTS_ONLY" (it was not fully implemented with ACES, and was lacking mid gray adjustments) -Polish code a bit and remove unnecesary old stuff

…ersion; fixed swapchain setting

… gamma 2.2

…eld-Native-HDR

… code base

Make DICE tonemap per channel (WIP)

…orRatio" having highly unstable values (likely due to tiny values dividing tiny values, producing unstable results).

… which only enables gamma 2.2 emulation after LUT application as the other mode is incorrect

…ensity towards the lower and top end remove old actual sigmoidal contrast adjustment as it is no longer needed

- add ability to clamp to AP0 for scRGB (currently disabled for testing; should be enabled for release) - split shader work for FP16 output and UNORM10 output

…from midtones (shadow/toe end) (disabled as it looks bad), and one to not start the DICE highlights tonemapper until hitting 33% of the screen peak brightness (enabled as it looks good under good HDR calibration settings)

… (we still clamp them away for scRGB on output)

…oken, it needs to have gamma applied and it's too complicated to do that in the HDR composite shader ("AdditionalNeutralLUTPercentage") can be used for that

…pace and apply the same fixes to the improved one

EndlesslyFlowering · 2023-10-18T04:12:01Z

-		const float filmGrainInvSize = 1.f / (FILM_GRAIN_TEXTURE_SIZE - 1u);
-		const float filmGrainHalfSize = 521.f; //TODO: rename in case we keep this as 521
+		static const float filmGrainInvSize = 1.f / (FILM_GRAIN_TEXTURE_SIZE - 1);
+		static const float filmGrainHalfSize = 511.f;


511 seems to be the correct value as: (1024 / 2) - 1 = 511
this value also produces the least artifacts

clshortfuse

In the improved version, as the code is, just removes nuance of grain completely:

Performing the requested changes achieves this look:

Without the PR, the master branch looks like this:

clshortfuse · 2023-10-18T04:45:47Z


 #if 0 // Output Visualization
-	float oldY = Luminance(GAMMA_TO_LINEAR(gammaColor));
+	float oldY = Luminance(GAMMA_TO_LINEAR(linearColor));


linearColor doesn't exist. This should probably be color now.

clshortfuse · 2023-10-18T04:56:24Z

+
+		float3 linearColor = inputColor;
+		if (isInOutColorLinear) {
+			linearColor /= PQMaxWhitePoint;


This line seems better removed since without it, the density no longer scales properly.

clshortfuse · 2023-10-18T04:56:44Z

+			outputColor = max(outputColor, 0.f);
+			outputColor = AP0D65_To_BT709(outputColor);
+
+			outputColor *= PQMaxWhitePoint;


If removed aboved, the should be removed here as well.

clshortfuse · 2023-10-18T04:57:53Z

-		;
+			* 3.333f; // Bump to 10%
+
+		yChange *= 0.0666f;


Needs a comment explaining it's usage (and can be folded into the above declaration. Though, after adjusting for 80 nits, it's better removed.

was just testing values that looked right ... 😓

clshortfuse · 2023-10-18T05:25:01Z

-
 		float yAdjustment = isHDR
-			? (HdrDllPluginConstants.HDRGamePaperWhiteNits / WhiteNits_sRGB)
+			? (HdrDllPluginConstants.HDRGamePaperWhiteNits / PQMaxWhitePoint)


Film density works in a range that is 0 to 3, whilch should be 0-1 in Y space. Middle gray matches RGB Luminance at ~18%. Anything above diffuse white is more or less the same as having no grain. That's why we want SDR to stay as it is, but HDR back down.

See:

the logarithm of light intensity is in foot candle in the source
so a density of 3 is achieved at ~3.6. so 10^3.6 = 3981 lm/ft² which is 42851 cd/m²!

not sure why we cap at mid gray then?

I get that it is supposed to cap
but film never has no grain?

density is based on exposure time and the light intensity

We're not capping at midgray, we're basing around 0.75 optical density (mid gray). FIlm Grain is a perceived thing, not a physical thing. It depends on different attributes. If there's no density (white on negative), there's no graininess. But at the same time, anything near black output (white on negative), has stronger perceived graininess. The formula for computed perceptual graininess is part of computeFilmGraininess. Once it hits around diffuse white, there's no real perceived grain (near ~0), so we can set bounds around paperwhite.

Right now there's no curve to match exposure to density (eg: film stock). It's just linear, but close enough for the intended effect.

The source code already links to the references.

Filoppi · 2023-10-18T20:59:31Z

Linear (SDR) until output is a branch. For now it's on but let's support both cases.

EndlesslyFlowering and others added 30 commits September 30, 2023 01:19

add readability change to Hable tone mapping

8c6f11e

Polish LUTs a bit, simplify math and fix some issues with it (mostly …

a739a25

…to constrain the output within 0-1)

WIP: Hable improvements

b5ddd42

Re-implement PR #4 from @clshortfuse

3d12aef

(with some minor changes)

avoid using defines in funcs

703c7de

Add:

8420ceb

-Inverse tonemap by luminance (to maintain SDR hue) -Inverse tonemap highlights only (to maintain SDR crushed shadow filmic look) -Automatically find paper white multiplier by checking how much mid gray shifted with the inverse tonemapper -Improve LUTs shadow invalid colors

Move LUTs application to be after inverse tonemapping, this allows LU…

26f9b6b

…Ts to have HDR/scRGB colors (beyond 0-1)

Small tonemapper and LUT fixes (from ShortFuse)

d6cb484

small cleanup + describe the pow(x, 2.2) for the toeLength parameter

287f216

add struct for hable parameters which are shared between tone mapping…

0d3b403

… and inverse tone mapping

convert everything connected to shaders to use tab indentation and ad…

025b1c1

…d ".editorconfig"

forgot to change something

b156b7d

-Add support for "INVERT_TONEMAP_HIGHLIGHTS_ONLY" to ACES, so that mo…

ecd44e7

…st of the image will have the same look as SDR -Improve LUTs shadow gradient -Remove AutoHDR (it was just a fun test) -Clean up code

added photo mode blocking

18a3dd7

moved to address library; updated the plugin to the newest template v…

e871228

…ersion; fixed swapchain setting

add sigmoidal inspired contrast adjustment

a3a9e47

add macro for comparing linearization of the LUTs with either sRGB or…

24dda4d

… gamma 2.2

refactored settings; sdr's swapchain is now RGB10A2

785f83d

Merge branch 'master' of https://github.com/EndlesslyFlowering/Starfi…

a124fd9

…eld-Native-HDR

Add saturation formula

a79c9d9

Remove useless "#if"

5f69098

Add new gamma correction setting and unify all the defines around the…

adb5032

… code base

Add shader/pipeline root signatures for later use and modification

9bee85e

Gamma correction fixes.

c7f80c2

Make DICE tonemap per channel (WIP)

Fix RestorePostProcess() sometimes going crazy due to "postProcessCol…

1b45d9a

…orRatio" having highly unstable values (likely due to tiny values dividing tiny values, producing unstable results).

rename CORRECT_LUT_GAMMA_CHANGES_TYPE to EMULATE_SDR_GAMMA_APPEARANCE…

4682b38

… which only enables gamma 2.2 emulation after LUT application as the other mode is incorrect

adjust sigmoidal inspired contrast adjustment to bias the contrastInt…

1d7a7df

…ensity towards the lower and top end remove old actual sigmoidal contrast adjustment as it is no longer needed

luts: use OKLCH for color boost

dead6b0

luts: use OKLCH for color boost (#7)

da019f8

EndlesslyFlowering and others added 17 commits October 17, 2023 06:04

HDRComposite: fix error in Dice tonemapping in ICtCp

2462275

Copy:

2d0a0a4

- add ability to clamp to AP0 for scRGB (currently disabled for testing; should be enabled for release) - split shader work for FP16 output and UNORM10 output

Fix up minor mistakes on copy shader

d51839f

Added ExtendGamut setting in game

bba38af

HDRComposite: remove negative Luminance check as it causes black dots…

573de34

… (we still clamp them away for scRGB on output)

remove all negative luminance checks as we clamp on output anyways

49a1d0b

Copy: enable AP0 clamp on scRGB output

94597b6

Filmgrain: fix blending issues caused by negative luminance

39bc0c3

Fixed up LUT gradient scale (it was always multipled by the LUT size)

7ab8718

Remove debug drawing of LUTs mixing with the neutral LUT as it was br…

29badee

…oken, it needs to have gamma applied and it's too complicated to do that in the HDR composite shader ("AdditionalNeutralLUTPercentage") can be used for that

Wrap up gamma correction code

b998ce4

Bump version to 0.3

f2e3762

Small comment

4677b73

Disable development settings for release (re-enable them locally pls)

ccb1c0b

feat(filmgrain): add FilmGrainCap option

03edccb

Filmgrain: fix native filmgrain by making it work in approximate PQ s…

e606b0f

…pace and apply the same fixes to the improved one

EndlesslyFlowering requested a review from clshortfuse October 18, 2023 04:06

EndlesslyFlowering commented Oct 18, 2023

View reviewed changes

clshortfuse requested changes Oct 18, 2023

View reviewed changes

EndlesslyFlowering force-pushed the master branch 3 times, most recently from dfe574b to 9bfe894 Compare November 11, 2023 19:57

EndlesslyFlowering force-pushed the master branch 3 times, most recently from dc5063a to 998d6df Compare November 29, 2023 18:44

EndlesslyFlowering force-pushed the big_grain_fix branch from 5f821f1 to e606b0f Compare November 29, 2023 19:40

EndlesslyFlowering force-pushed the master branch 2 times, most recently from eca7734 to 43eed47 Compare November 29, 2023 20:51

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Filmgrain: fix native filmgrain by making it work in approximate PQ space and apply the same fixes to the improved one#21

Filmgrain: fix native filmgrain by making it work in approximate PQ space and apply the same fixes to the improved one#21
EndlesslyFlowering wants to merge 287 commits into
masterfrom
big_grain_fix

EndlesslyFlowering commented Oct 18, 2023 •

edited

Loading

Uh oh!

EndlesslyFlowering Oct 18, 2023

Uh oh!

clshortfuse left a comment •

edited

Loading

Uh oh!

clshortfuse Oct 18, 2023

Uh oh!

clshortfuse Oct 18, 2023

Uh oh!

clshortfuse Oct 18, 2023

Uh oh!

clshortfuse Oct 18, 2023

Uh oh!

EndlesslyFlowering Oct 19, 2023

Uh oh!

clshortfuse Oct 18, 2023 •

edited

Loading

Uh oh!

EndlesslyFlowering Oct 19, 2023 •

edited

Loading

Uh oh!

EndlesslyFlowering Oct 19, 2023

Uh oh!

clshortfuse Oct 19, 2023

Uh oh!

Filoppi commented Oct 18, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

EndlesslyFlowering commented Oct 18, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

clshortfuse left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

clshortfuse Oct 18, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

EndlesslyFlowering Oct 19, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Filoppi commented Oct 18, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

EndlesslyFlowering commented Oct 18, 2023 •

edited

Loading

clshortfuse left a comment •

edited

Loading

clshortfuse Oct 18, 2023 •

edited

Loading

EndlesslyFlowering Oct 19, 2023 •

edited

Loading