Color picker malfunctions after manually entering hex value with full red channel #115562

New Issue

Sun Kim · 2023-11-29T07:36:52+01:00

Sun Kim commented

2023-11-29 07:36:52 +01:00

System Information
Operating system: Windows-10-10.0.19045-SP0 64 Bits
Graphics card: NVIDIA GeForce RTX 2060 SUPER/PCIe/SSE2 NVIDIA Corporation 4.6.0 NVIDIA 546.17

Blender Version
Broken: version: 4.1.0 Alpha, branch: main, commit date: 2023-11-28 21:58, hash: 09caf5969010
Worked: 3.6

Caused by 1bcf31a874

Short description of error
Color picker malfunctions after manually entering a hex value with full red channel (#FFxxxx).

As far as I can see this happens in material, world, geometry nodes modifier panels, and pickers inside nodes. It does not happen with pickers in the sculpt/vertex/texture paint tools and bone display color selection. I guess this only happens with pickers in user-created data?

Once the picker is broken it stays broken for that input, for that file, even after re-launching Blender and loading the file.

Here is the file with broken pickers: color_picker_issue.blend

Also, while the pickers don't break in 3.6, loading the broken file in 3.6 shows the same broken pickers.

**System Information** Operating system: Windows-10-10.0.19045-SP0 64 Bits Graphics card: NVIDIA GeForce RTX 2060 SUPER/PCIe/SSE2 NVIDIA Corporation 4.6.0 NVIDIA 546.17 **Blender Version** Broken: version: 4.1.0 Alpha, branch: main, commit date: 2023-11-28 21:58, hash: `09caf5969010` Worked: 3.6 Caused by 1bcf31a874 **Short description of error** Color picker malfunctions after manually entering a hex value with full red channel (#FFxxxx). <video src="/attachments/f95d3175-e58d-425b-931f-8b5f5e0cb091" title="color_picker_issue.mp4" controls></video> As far as I can see this happens in material, world, geometry nodes modifier panels, and pickers inside nodes. It does not happen with pickers in the sculpt/vertex/texture paint tools and bone display color selection. I guess this only happens with pickers in user-created data? Once the picker is broken it stays broken for that input, for that file, even after re-launching Blender and loading the file. Here is the file with broken pickers: [color_picker_issue.blend](/attachments/9a39dace-80e9-42e1-907d-dc2848a54bd8) Also, while the pickers don't break in 3.6, loading the broken file in 3.6 shows the same broken pickers.

color_picker_issue.mp4

1.6 MiB

color_picker_issue.blend

1011 KiB

Sun Kim added the

labels 2023-11-29 07:36:53 +01:00

Pratik Borhade commented

2023-11-30 10:03:46 +01:00

Hi, can confirm. Can not bisect atm (don't have intermediate build right now to narrow down good-bad commits)

Pratik Borhade added

and removed

labels 2023-11-30 10:04:00 +01:00

Philipp Oeser commented

2023-11-30 10:46:24 +01:00

Will check on this

Philipp Oeser commented

2023-11-30 12:49:49 +01:00

Caused by 1bcf31a874

CC @Sergey
CC @Eary

Caused by 1bcf31a874fb4e9444bf66e43e16fa58dcee20bf CC @Sergey CC @Eary

Zijun Zhou commented

2023-11-30 19:11:45 +01:00

The reason seems to be the precision in calculating the "XYZ D65 to Lin709" matrix.

Replace in the config the Linear Rec.709 stanza with this should solve the issue:

  - !<ColorSpace>
    name: Linear Rec.709
    aliases: [Linear, Linear BT.709, Linear BT.709 I-D65, Linear Tristimulus, linrec709, Utility - Linear - sRGB, Utility - Linear - Rec.709, lin_srgb, Linear Rec.709 (sRGB), lin_rec709_srgb, lin_rec709, lin_srgb, "CGI: Linear - Rec.709"]
    family: Linear
    equalitygroup:
    bitdepth: 32f
    description: |
      Linear BT.709 with illuminant D65 white point
    isdata: false
    from_scene_reference: !<GroupTransform>
      children:
        - !<ColorSpaceTransform> {src: Linear CIE-XYZ E, dst: Linear CIE-XYZ D65}
        - !<MatrixTransform> {matrix: [3.2404373564920070, -1.5371305409000549, -0.4985288240756933, 0, -0.9692674982687597, 1.8760136862977035, 0.0415560804587997, 0, 0.0556434463874256, -0.2040259700872272, 1.0572254813610864, 0, 0, 0, 0, 1]}

The reason seems to be the precision in calculating the "XYZ D65 to Lin709" matrix. Replace in the config the Linear Rec.709 stanza with this should solve the issue: ``` - !<ColorSpace> name: Linear Rec.709 aliases: [Linear, Linear BT.709, Linear BT.709 I-D65, Linear Tristimulus, linrec709, Utility - Linear - sRGB, Utility - Linear - Rec.709, lin_srgb, Linear Rec.709 (sRGB), lin_rec709_srgb, lin_rec709, lin_srgb, "CGI: Linear - Rec.709"] family: Linear equalitygroup: bitdepth: 32f description: | Linear BT.709 with illuminant D65 white point isdata: false from_scene_reference: !<GroupTransform> children: - !<ColorSpaceTransform> {src: Linear CIE-XYZ E, dst: Linear CIE-XYZ D65} - !<MatrixTransform> {matrix: [3.2404373564920070, -1.5371305409000549, -0.4985288240756933, 0, -0.9692674982687597, 1.8760136862977035, 0.0415560804587997, 0, 0.0556434463874256, -0.2040259700872272, 1.0572254813610864, 0, 0, 0, 0, 1]} ```

Sergey Sharybin commented

2023-12-01 17:00:15 +01:00

The change seems to solve the issue on user level.
Not sure how to go about revieweing the change though. Is there some known ground-truth matrix for this?

@brecht @nathanvegdahl Do you mind having a look here, because it will be nice to solve for 4.0.2 and I might be away for that.

The change seems to solve the issue on user level. Not sure how to go about revieweing the change though. Is there some known ground-truth matrix for this? @brecht @nathanvegdahl Do you mind having a look here, because it will be nice to solve for 4.0.2 and I might be away for that.

Zijun Zhou commented

2023-12-01 20:29:12 +01:00

Is there some known ground-truth matrix for this?

Not that I know of.

The key point seems to be the D65 coordinate.

This is the tool I used for generating the XYZ to RGB matrix:
https://observablehq.com/@danburzo/color-matrix-calculator

It comes with a D65 xy coordinate of (x: 0.31271, y: 0.32902) , and this is also the widely cited D65 coordinate.

But now I looked into it, different sources seems to disagree on the actual coordinate of D65:

https://en.wikipedia.org/wiki/Talk:Standard_illuminant#Series_D

All of these seems super complex and controversial, and is beyond my math ability to understand.

All I get from this is that, the D65 coordinate doesn't have a single ground-truth value, it depends on the method they calculated it, as well as some different way of rounding different values in different steps that they do.

The new matrix from my above comment took the D65 XYZ coordinate from this site:
http://www.brucelindbloom.com/index.html?Eqn_ChromAdapt.html

(I have varified the adaptation matrices listed in the site with this script so I trust this source)

X = 0.95047, Y = 1, Z = 1.08883

I ran it through the XYZ to xyY calculator:
http://www.brucelindbloom.com/index.html?ColorCalculator.html

The result is x = 0.312727, y= 0.329023

Then I pasted the result value back to the RGB matrix generator tool, and got the above matrix I posted. And it seems to work.

> Is there some known ground-truth matrix for this? Not that I know of. The key point seems to be the D65 coordinate. This is the tool I used for generating the XYZ to RGB matrix: https://observablehq.com/@danburzo/color-matrix-calculator It comes with a D65 xy coordinate of (x: 0.31271, y: 0.32902) , and this is also the widely cited D65 coordinate. But now I looked into it, different sources seems to disagree on the actual coordinate of D65: https://en.wikipedia.org/wiki/Talk:Standard_illuminant#Series_D All of these seems super complex and controversial, and is beyond my math ability to understand. All I get from this is that, the D65 coordinate doesn't have a single ground-truth value, it depends on the method they calculated it, as well as some different way of rounding different values in different steps that they do. The new matrix from my above comment took the D65 XYZ coordinate from this site: http://www.brucelindbloom.com/index.html?Eqn_ChromAdapt.html (I have varified the adaptation matrices listed in the site with [this script](https://github.com/EaryChow/Chromatic-Adaptation-Matrix-Generator/blob/main/chromatic_adaptation_matrix_generator.py) so I trust this source) X = 0.95047, Y = 1, Z = 1.08883 I ran it through the XYZ to xyY calculator: http://www.brucelindbloom.com/index.html?ColorCalculator.html The result is x = 0.312727, y= 0.329023 Then I pasted the result value back to the RGB matrix generator tool, and got the above matrix I posted. And it seems to work.

Brecht Van Lommel referenced a pull request that will close this issue

2023-12-04 18:39:52 +01:00

Fix #115562: Issues with color picker and color values near 1 #115767

Brecht Van Lommel referenced this issue from a commit

2023-12-04 20:06:09 +01:00

Fix #115562: Issues with color picker and color values near 1

Brecht Van Lommel closed this issue

2023-12-04 20:06:10 +01:00

Blender Bot added

Status

Resolved

and removed