GLSL - color conversion from RGB to Value uses unweighted average (like vector conversion) #45674

New Issue

Michael Otto · 2015-08-04T12:24:48+02:00

Michael Otto commented

2015-08-04 12:24:48 +02:00

System Information
windows, gtx980

Blender Version
2.75
Worked: (optional)
2.74

Short description of error
When a node with float input (like a math node) is connected to a color, the color was converted using a weighted average based on human perception:

f = dot(col, vec3(0.35, 0.45, 0.2)) until 2.74,

Now, it's

(col.r, col.g, col.b) / 3.0

no matter if a vector or color is connected to the float input.

Exact steps for others to reproduce the error

Attached blend file: export a shader in 2.74 and 2.75, (gpu.export_shader) and diff them
In 2.74, the add node function is clalled by sth like

math_add(dot(tmp2.rgb, vec3(0.35, 0.45, 0.2)), dot(tmp4.rgb, vec3(0.35, 0.45, 0.2)), tmp7);

in 2.75, it's

math_add((tmp2.r + tmp2.g + tmp2.b) / 3.0, (tmp4.r + tmp4.g + tmp4.b) / 3.0, tmp7);

color_value_conversion.blend

**System Information** windows, gtx980 **Blender Version** 2.75 Worked: (optional) 2.74 **Short description of error** When a node with float input (like a math node) is connected to a color, the color was converted using a weighted average based on human perception: f = dot(col, vec3(0.35, 0.45, 0.2)) until 2.74, Now, it's (col.r, col.g, col.b) / 3.0 no matter if a vector or color is connected to the float input. **Exact steps for others to reproduce the error** Attached blend file: export a shader in 2.74 and 2.75, (gpu.export_shader) and diff them In 2.74, the add node function is clalled by sth like math_add(dot(tmp2.rgb, vec3(0.35, 0.45, 0.2)), dot(tmp4.rgb, vec3(0.35, 0.45, 0.2)), tmp7); in 2.75, it's math_add((tmp2.r + tmp2.g + tmp2.b) / 3.0, (tmp4.r + tmp4.g + tmp4.b) / 3.0, tmp7); [color_value_conversion.blend](https://archive.blender.org/developer/F219036/color_value_conversion.blend)

Michael Otto commented

2015-08-04 12:24:48 +02:00

Changed status to: 'Open'

Michael Otto commented

2015-08-04 12:24:48 +02:00

Added subscriber: @MichaelOtto

Sebastian Ullrich commented

2015-08-04 12:46:03 +02:00

Added subscriber: @souljedi

tyoc213 commented

2015-08-04 14:33:16 +02:00

Added subscriber: @tyoc213

Kévin Dietrich commented

2015-08-05 03:31:43 +02:00

Added subscriber: @kevindietrich

Kévin Dietrich commented

2015-08-05 03:31:43 +02:00

Changed status from 'Open' to: 'Archived'

Kévin Dietrich closed this issue

2015-08-05 03:31:43 +02:00

Kévin Dietrich self-assigned this 2015-08-05 03:31:43 +02:00

Kévin Dietrich commented

2015-08-05 03:31:43 +02:00

This is not a bug, rather, it's a fix. There was an inconsistency between how Blender internal render engine and GLSL handled color to float socket conversion, which was eventually fixed by 4139686817. Doing a weighting based on human perception should be carefully handled by the color management pipeline and has nothing to do with such conversion.

Thanks for the report, but again this is not bug.

This is not a bug, rather, it's a fix. There was an inconsistency between how Blender internal render engine and GLSL handled color to float socket conversion, which was eventually fixed by 4139686817. Doing a weighting based on human perception should be carefully handled by the color management pipeline and has nothing to do with such conversion. Thanks for the report, but again this is not bug.

Michael Otto commented

2015-08-05 10:41:27 +02:00

Makes sense. Thank you for looking into it!

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

GLSL - color conversion from RGB to Value uses unweighted average (like vector conversion) #45674