Cycles: Rework component layering in Principled BSDF #110864

Merged

Lukas Stockner merged 10 commits from LukasStockner/blender:layering-principled into main

2023-08-10 23:53:44 +02:00

Lukas Stockner commented

2023-08-07 01:29:08 +02:00

Member

Overall, this PR reworks the component layering in the Principled BSDF in order to ensure that energy is preserved and conserved.

This includes:

Implementing support for the OSL layer() function
Implementing albedo estimation for some of the closures (for layering purposes)
- The specular layer that the Principled BSDF uses has a proper tabulated albedo lookup, the others are still approximations
Removing the custom "Principled Diffuse" and replacing it with the classic lambertian Diffuse, since the layering logic takes care of energy now

Note that this changes the look of the Principled BSDF noticeably in some cases, but that's needed, since the cases where it looks different are the ones that strongly violate energy conservation (mostly grazing reflections with strong Specular).

Overall, this PR reworks the component layering in the Principled BSDF in order to ensure that energy is preserved and conserved. This includes: - Implementing support for the OSL `layer()` function - Implementing albedo estimation for some of the closures (for layering purposes) - The specular layer that the Principled BSDF uses has a proper tabulated albedo lookup, the others are still approximations - Removing the custom "Principled Diffuse" and replacing it with the classic lambertian Diffuse, since the layering logic takes care of energy now Note that this changes the look of the Principled BSDF noticeably in some cases, but that's needed, since the cases where it looks different are the ones that strongly violate energy conservation (mostly grazing reflections with strong Specular).

👍 5 ❤️ 6 🚀 2

Lukas Stockner added the

Render & Cycles

label 2023-08-07 01:29:08 +02:00

Brecht Van Lommel commented

2023-08-08 18:25:31 +02:00

Owner

@blender-bot package

@blender-bot package

Blender Bot commented

2023-08-08 18:25:35 +02:00

Member

Package build started. Download here when ready.

Package build started. [Download here](https://builder.blender.org/download/patch/PR110864) when ready.

Brecht Van Lommel reviewed 2023-08-08 18:33:58 +02:00

Brecht Van Lommel left a comment

Owner

I have not checked every line in detail, but the overall approach looks great to me.

Some images and timings would be nice.

I have not checked every line in detail, but the overall approach looks great to me. Some images and timings would be nice.

intern/cycles/kernel/osl/osl.h Outdated

						
				@ -71,2 +71,4 @@

				  float3 weight_stack[16];

				  ccl_private const OSLClosure *closure_stack[16];

				  int layer_stack_level = -1;

				  float3 layer_albedo;

Brecht Van Lommel commented

2023-08-08 18:28:23 +02:00

Owner

Typically better for the GPU compiler to initialize this.

Typically better for the GPU compiler to initialize this.

brecht marked this conversation as resolved

Lukas Stockner commented

2023-08-09 07:34:17 +02:00

Author

Member

Some images and timings would be nice.

Scene	Old	New	New with exposure adjusted
Diorama	1m25s	1m24s	/
3.5 Splash	2m48s	2m49s	2m49s, exposure +0.2
2.81 Splash	55s	56s	56s, exposure +0.2

> Some images and timings would be nice. ------ | Scene | Old | New | New with exposure adjusted | | - | - | - | - | | Diorama | ![diorama_old.png](/attachments/cb3c3f11-27ba-488b-a896-a0b10b933ae4) 1m25s | ![diorama_new.png](/attachments/ab54ab65-f250-4bd9-b309-ee8ac1826f31) 1m24s | / | | 3.5 Splash | ![splash35_old.png](/attachments/9e1e6ede-4b2e-4e11-983f-b94bd1f296c2) 2m48s | ![splash35_new.png](/attachments/8e38e468-2b44-486a-9636-c97a1ccc0273) 2m49s | ![splash35_new_exposure0.2.png](/attachments/c332889c-500e-4041-a2d1-06bcd2167e03) 2m49s, exposure +0.2 | | 2.81 Splash | ![splash281_old.png](/attachments/1a0a84f7-dea6-4386-860d-3b85323d6d2e) 55s | ![splash281_new.png](/attachments/ed2dc8a8-edae-492c-83cb-e9a007096446) 56s | ![splash281_new_exposure0.2.png](/attachments/d6530f6e-6c42-4e8f-b682-6b38b4130e9a) 56s, exposure +0.2 |

diorama_new.png

438 KiB

splash35_new.png

1.7 MiB

splash281_new_exposure0.2.png

581 KiB

diorama_old.png

395 KiB

splash35_old.png

1.7 MiB

splash35_new_exposure0.2.png

1.7 MiB

splash281_old.png

576 KiB

splash281_new.png

569 KiB

Brecht Van Lommel requested changes 2023-08-09 15:14:17 +02:00

Brecht Van Lommel left a comment

Owner

Render time and look differences seem within an acceptable range.

In the regression tests, render passes specular direct seems wrong, there is a checker texture visible in that pass.

Render time and look differences seem within an acceptable range. In the regression tests, `render passes specular direct` seems wrong, there is a checker texture visible in that pass.

intern/cycles/kernel/osl/osl.h Outdated

						
				@ -114,0 +141,4 @@

				        /* We just finished processing the top layers of a Layer closure, so adjust the weight to

				         * account for the layering. */

				        weight *= saturatef(1.0f - reduce_max(layer_albedo / weight));

				        /* TODO: Skip if weight zero */

Brecht Van Lommel commented

2023-08-09 15:13:23 +02:00

Owner

Is this still intended to be resolved?

Is this still intended to be resolved?

Lukas Stockner commented

2023-08-10 02:46:04 +02:00

Author

Member

Yep, resolved now. I had tried to find a way to not repeat the stack popping code, but didn't come up with anything.

Yep, resolved now. I had tried to find a way to not repeat the stack popping code, but didn't come up with anything.

brecht marked this conversation as resolved

Lukas Stockner commented

2023-08-10 02:44:54 +02:00

Author

Member

In the regression tests, render passes specular direct seems wrong, there is a checker texture visible in that pass.

I've improved the color pass now by extending and improving the precomputed tables that are used to lookup the expected Fresnel effect depending on roughness, instead of just using the smooth surface approximation.

With this change, the checkerboard is completely gone when using the material in a "sphere in front of white background" setup.

However, in the test scene it's still slightly visible. This is expected - bounce rays at grazing angles are occluded by the back wall, while more vertical rays aren't. And since the Fresnel effect depends on angle, the effective mix between F0 (checkerboard) and F90 (white) changes, and so the precomputed value no longer fits perfectly.

In case you're wondering how this worked before the change (as I did): The material has Specular set to zero, and the old formulation of the Principled BSDF had a weird quirk where the Specular input affected the Fresnel curve that was used for metals. By setting it to zero, Fresnel was effectively disabled. If you increase Specular, you also get the checkerboard in old builds.

> In the regression tests, render passes specular direct seems wrong, there is a checker texture visible in that pass. I've improved the color pass now by extending and improving the precomputed tables that are used to lookup the expected Fresnel effect depending on roughness, instead of just using the smooth surface approximation. With this change, the checkerboard is completely gone when using the material in a "sphere in front of white background" setup. However, in the test scene it's still slightly visible. This is expected - bounce rays at grazing angles are occluded by the back wall, while more vertical rays aren't. And since the Fresnel effect depends on angle, the effective mix between F0 (checkerboard) and F90 (white) changes, and so the precomputed value no longer fits perfectly. In case you're wondering how this worked before the change (as I did): The material has Specular set to zero, and the old formulation of the Principled BSDF had a weird quirk where the Specular input affected the Fresnel curve that was used for metals. By setting it to zero, Fresnel was effectively disabled. If you increase Specular, you also get the checkerboard in old builds.

Lukas Stockner referenced this pull request

2023-08-10 08:24:28 +02:00

Cycles: Tweak Principled BSDF Subsurface parameters #110989

~~Lukas Stockner referenced this pull request 2023-08-10 10:23:31 +02:00~~

Cycles: Rework Principled BSDF Clearcoat #110993

Brecht Van Lommel approved these changes 2023-08-10 18:36:32 +02:00

Brecht Van Lommel left a comment

Owner

Thanks, looks good to me now.

Thanks, looks good to me now.

Lukas Stockner force-pushed layering-principled from 86100d921b to 6a3c2f62c5

2023-08-10 23:51:07 +02:00

Compare

Lukas Stockner merged commit 2ac0b36e4e into main

2023-08-10 23:53:44 +02:00

Lukas Stockner deleted branch layering-principled

2023-08-10 23:53:46 +02:00

Lukas Stockner referenced this issue from a commit

2023-08-10 23:53:46 +02:00

Cycles: Rework component layering in Principled BSDF

Lukas Stockner referenced this pull request

2023-08-11 10:07:19 +02:00

Cycles: Tweak Principled BSDF Subsurface parameters #110989

Christophe Hery referenced this pull request

2023-08-11 16:58:52 +02:00

Cycles: Tweak Principled BSDF Subsurface parameters #110989

Brecht Van Lommel referenced this pull request

2023-08-14 16:38:17 +02:00

Principled v2 BSDF #99447

Brecht Van Lommel referenced this issue from a commit

2023-12-05 01:37:12 +01:00

Rework component layering in principled bsdf

Sign in to join this conversation.

No reviewers

No Label

Animation & Rigging

Asset Browser Project Overview

Automated Testing

Blender Asset Bundle

Dependency Graph

Development Management

EEVEE & Viewport

Images & Movies

Motion Tracking

Nodes & Physics

Pipeline, Assets & IO

Platforms, Builds & Tests

Render & Cycles

Render Pipeline

Sculpt, Paint & Texture

Video Sequencer

Virtual Reality

Blender 2.8 Project

Milestone 1: Basic, Local Asset Browser

Good First Issue

Animation & Rigging

Development Management

EEVEE & Viewport

Nodes & Physics

Pipeline, Assets & IO

Platforms, Builds & Tests

Render & Cycles

Sculpt, Paint & Texture

Needs Info from Developers

Needs Information from User

No Milestone

No project

No Assignees

3 Participants

Notifications

Due Date

The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: blender/blender#110864

No description provided.