OpenColorIO 2.0 Upgrade #84819

Closed
opened 2021-01-18 13:26:44 +01:00 by Brecht Van Lommel · 37 comments
  • Code changes to make Blender build
  • Configuration file update
  • Update precompiled libraries (#83246 (VFX Reference Platform 2021 Compatibility))
  • Use new features
    • Update configuration to use new features and useful builtin transforms
    • Support menu categories for long lists of color space
    • Check for performance improvements (especially with OpenGL render)
- [x] Code changes to make Blender build - [x] Configuration file update - [x] Update precompiled libraries (#83246 (VFX Reference Platform 2021 Compatibility)) - [ ] Use new features - [ ] Update configuration to use new features and useful builtin transforms - [ ] Support menu categories for long lists of color space - [ ] Check for performance improvements (especially with OpenGL render)
Author
Owner

Changed status from 'Needs Triage' to: 'Confirmed'

Changed status from 'Needs Triage' to: 'Confirmed'
Author
Owner

Added subscribers: @Jeroen-Bakker, @brecht

Added subscribers: @Jeroen-Bakker, @brecht
Member

Just to give a status of the branch.

The smaller changes have been done.
ocio wrapper in intern should be updated to use the ocio pipeline. I didn't see a way around this API change. In the tmp-ocio-v2 I was planning for these changes but haven't done any code changes.

Just to give a status of the branch. The smaller changes have been done. ocio wrapper in intern should be updated to use the ocio pipeline. I didn't see a way around this API change. In the tmp-ocio-v2 I was planning for these changes but haven't done any code changes.
Brecht Van Lommel changed title from OpenColor 2.0 Upgrade to OpenColorIO 2.0 Upgrade 2021-01-18 14:35:33 +01:00
Member

Added subscriber: @Blendify

Added subscriber: @Blendify

Added subscriber: @baoyu

Added subscriber: @baoyu

Since blender integrated OCIO, there's no way to set RRT and ODT seperately. RRT and ODT are tied together and can't be changed to any useful transforms other than srgb.
Could we have full ACES support or at least fulfill: 1 ability to seperately set IDT, RRT and ODT; 2 more useful transforms like Display P.3, Rec 2020 etc. and a correct preview, output, compositor tools that respects the color management.

Since blender integrated OCIO, there's no way to set RRT and ODT seperately. RRT and ODT are tied together and can't be changed to any useful transforms other than srgb. Could we have full ACES support or at least fulfill: 1 ability to seperately set IDT, RRT and ODT; 2 more useful transforms like Display P.3, Rec 2020 etc. and a correct preview, output, compositor tools that respects the color management.

Added subscriber: @svenstaro

Added subscriber: @svenstaro

Hey guys, in Arch Linux we're currently in the middle of the opencolorio 2.0 rebuild and Blender is currently blocking these efforts. Could you perhaps give me an update on this?

Hey guys, in Arch Linux we're currently in the middle of the opencolorio 2.0 rebuild and Blender is currently blocking these efforts. Could you perhaps give me an update on this?

Added subscriber: @Memento

Added subscriber: @Memento

This issue was referenced by 1b4961b318

This issue was referenced by 1b4961b318f14064bc3c915da7206a74146af95d

This issue was referenced by 6b40ee608c

This issue was referenced by 6b40ee608c5a1dbe67870fab4b83cce4cc4b54f5
Author
Owner

I have a set of patches now for the upgrade, they still need to go through review.

There is no specific timeline for when this lands, I would guess somewhere in the next two weeks.

I have a set of patches now for the upgrade, they still need to go through review. There is no specific timeline for when this lands, I would guess somewhere in the next two weeks.

Cool, thanks for the fast action. :)

For the time being, we've added an opencolorio1 package that Blender builds against but I'd like to drop that package again ASAP.

Cool, thanks for the fast action. :) For the time being, we've added an opencolorio1 package that Blender builds against but I'd like to drop that package again ASAP.

Added subscriber: @SteffenD

Added subscriber: @SteffenD
Author
Owner

Added subscriber: @troy_s

Added subscriber: @troy_s
Author
Owner

@troy_s, is there a plan to update Filmic for OpenColorIO 2.0?

OpenColorIO has builtin transform for ACES and various displays now, and they work based on the new display reference space concept, using CIE-XYZ [D65](https://archive.blender.org/developer/D65).
https://opencolorio.readthedocs.io/en/latest/configurations/ocio_v2_demo.html

I'd like to add more displays and the ACES RRT to our config, and later figure out how we can also ship a config with ACES refererence space.

However lookup tables like filmic_to_0-70_1-03.spi1d I guess contain both part of the view transform and display transform, and it's not clear to me how they were constructed and how they might be split.

@troy_s, is there a plan to update Filmic for OpenColorIO 2.0? OpenColorIO has builtin transform for ACES and various displays now, and they work based on the new display reference space concept, using `CIE-XYZ [D65](https://archive.blender.org/developer/D65)`. https://opencolorio.readthedocs.io/en/latest/configurations/ocio_v2_demo.html I'd like to add more displays and the ACES RRT to our config, and later figure out how we can also ship a config with ACES refererence space. However lookup tables like `filmic_to_0-70_1-03.spi1d` I guess contain both part of the view transform and display transform, and it's not clear to me how they were constructed and how they might be split.

is there a plan to update Filmic for OpenColorIO 2.0?

Yes. I was letting the dust settle on v2.0 first.

OpenColorIO has builtin transform for ACES and various displays now, and they work based on the new display reference space concept, using CIE-XYZ D65.

Indeed. They came around to that idea finally, but sadly left out the display colourimetry aspect that would permit DCCs to query the chosen display colourimetry. It’s relevant to Blender due to the need to manage colour wheels, gradients, and composite in display linear. Might help to formulate a plan and have your voice there in the nearer future.

I'd like to add more displays and the ACES RRT to our config, and later figure out how we can also ship a config with ACES refererence space.

Should be rather trivial, but perhaps waiting for v2 might make more sense here. ACES brings so many broken facets to the table that people without an in-house colour scientist or colorist would likely be left helpless to repair things. ACES use is more or less as simple as a download, so not certain as to how much “better” it can get than the canonized sources.

The most prominent gains Blender can make is to get the UI up to a managed state. Integration with wider gamut spaces prior to that will be problematic.

Consider the output display encoded for BT.709 based displays. It is trivial to pull the inverse EOTF out of the lookups, and keep them display linear “clean”. It might be worth discussing how this can fit in nicely with Clem’s more recent display linear manipulations for overlays etc.

Available on Rocket if need be.

>is there a plan to update Filmic for OpenColorIO 2.0? Yes. I was letting the dust settle on v2.0 first. > OpenColorIO has builtin transform for ACES and various displays now, and they work based on the new display reference space concept, using CIE-XYZ [D65](https://archive.blender.org/developer/D65). Indeed. They came around to that idea finally, but sadly left out the display colourimetry aspect that would permit DCCs to query the chosen display colourimetry. It’s relevant to Blender due to the need to manage colour wheels, gradients, and composite in display linear. Might help to formulate a plan and have your voice there in the nearer future. > I'd like to add more displays and the ACES RRT to our config, and later figure out how we can also ship a config with ACES refererence space. Should be rather trivial, but perhaps waiting for v2 might make more sense here. ACES brings so many broken facets to the table that people without an in-house colour scientist or colorist would likely be left helpless to repair things. ACES use is more or less as simple as a download, so not certain as to how much “better” it can get than the canonized sources. The most prominent gains Blender can make is to get the UI up to a managed state. Integration with wider gamut spaces prior to that will be problematic. Consider the output display encoded for BT.709 based displays. It is trivial to pull the inverse EOTF out of the lookups, and keep them display linear “clean”. It might be worth discussing how this can fit in nicely with Clem’s more recent display linear manipulations for overlays etc. Available on Rocket if need be.
Author
Owner

In #84819#1108851, @troy_s wrote:
Indeed. They came around to that idea finally, but sadly left out the display colourimetry aspect that would permit DCCs to query the chosen display colourimetry. It’s relevant to Blender due to the need to manage colour wheels, gradients, and composite in display linear. Might help to formulate a plan and have your voice there in the nearer future.

As far as I know we have enough information from the config for handling color wheels, gradients, and compositing of UI elements. We really just need to do some operations in an approximately perceptually linear space, independent of the chosen display. And for that we can use cie_xyz_d65_interchange + some transform of our choice to a perceptually linear space.

Should be rather trivial, but perhaps waiting for v2 might make more sense here. ACES brings so many broken facets to the table that people without an in-house colour scientist or colorist would likely be left helpless to repair things. ACES use is more or less as simple as a download, so not certain as to how much “better” it can get than the canonized sources.

I will probably at the ACES RRT for the sRGB display since it fits in fine, and similar functionality is available in other apps.

The most prominent gains Blender can make is to get the UI up to a managed state. Integration with wider gamut spaces prior to that will be problematic.

Consider the output display encoded for BT.709 based displays. It is trivial to pull the inverse EOTF out of the lookups, and keep them display linear “clean”. It might be worth discussing how this can fit in nicely with Clem’s more recent display linear manipulations for overlays etc.

The displays are indeed less useful without more changes in Blender, though for file output it's still helpful. Probably will leave them out for now as long as they are not working with Filmic.

> In #84819#1108851, @troy_s wrote: > Indeed. They came around to that idea finally, but sadly left out the display colourimetry aspect that would permit DCCs to query the chosen display colourimetry. It’s relevant to Blender due to the need to manage colour wheels, gradients, and composite in display linear. Might help to formulate a plan and have your voice there in the nearer future. As far as I know we have enough information from the config for handling color wheels, gradients, and compositing of UI elements. We really just need to do some operations in an approximately perceptually linear space, independent of the chosen display. And for that we can use `cie_xyz_d65_interchange` + some transform of our choice to a perceptually linear space. > Should be rather trivial, but perhaps waiting for v2 might make more sense here. ACES brings so many broken facets to the table that people without an in-house colour scientist or colorist would likely be left helpless to repair things. ACES use is more or less as simple as a download, so not certain as to how much “better” it can get than the canonized sources. I will probably at the ACES RRT for the sRGB display since it fits in fine, and similar functionality is available in other apps. > The most prominent gains Blender can make is to get the UI up to a managed state. Integration with wider gamut spaces prior to that will be problematic. > > Consider the output display encoded for BT.709 based displays. It is trivial to pull the inverse EOTF out of the lookups, and keep them display linear “clean”. It might be worth discussing how this can fit in nicely with Clem’s more recent display linear manipulations for overlays etc. The displays are indeed less useful without more changes in Blender, though for file output it's still helpful. Probably will leave them out for now as long as they are not working with Filmic.

As far as I know we have enough information from the config for handling color wheels, gradients, and compositing of UI elements. We really just need to do some operations in an approximately perceptually linear space, independent of the chosen display. And for that we can use cie_xyz_d65_interchange + some transform of our choice to a perceptually linear space.

Simple case would be a changing of transfer function, or flipping a wheel to reveal greyscale; the assumption of invertibility doesn’t work across a full colour transform necessarily.

In the case of differing display colourimetry for UI, it is required to have the transfer function characteristic available.

I will probably at the ACES RRT for the sRGB display since it fits in fine, and similar functionality is available in other apps.

Seems problematic given the working spaces are different?

> As far as I know we have enough information from the config for handling color wheels, gradients, and compositing of UI elements. We really just need to do some operations in an approximately perceptually linear space, independent of the chosen display. And for that we can use cie_xyz_d65_interchange + some transform of our choice to a perceptually linear space. Simple case would be a changing of transfer function, or flipping a wheel to reveal greyscale; the assumption of invertibility doesn’t work across a full colour transform necessarily. In the case of differing display colourimetry for UI, it is required to have the transfer function characteristic available. > I will probably at the ACES RRT for the sRGB display since it fits in fine, and similar functionality is available in other apps. Seems problematic given the working spaces are different?
Author
Owner

In #84819#1108971, @troy_s wrote:
Simple case would be a changing of transfer function, or flipping a wheel to reveal greyscale; the assumption of invertibility doesn’t work across a full colour transform necessarily.

In the case of differing display colourimetry for UI, it is required to have the transfer function characteristic available.

Maybe we're talking about different things, but as far as I know we only need our chosen perceptually linear space to be invertible. Converting to the actual display space would always be in one direction only, at the very end.

Seems problematic given the working spaces are different?

Not sure how, going from a smaller to a wider gamut working space doesn't lose any information. It just means you're not taking advantage of the wider gamut.

> In #84819#1108971, @troy_s wrote: > Simple case would be a changing of transfer function, or flipping a wheel to reveal greyscale; the assumption of invertibility doesn’t work across a full colour transform necessarily. > > In the case of differing display colourimetry for UI, it is required to have the transfer function characteristic available. Maybe we're talking about different things, but as far as I know we only need our chosen perceptually linear space to be invertible. Converting to the actual display space would always be in one direction only, at the very end. > Seems problematic given the working spaces are different? Not sure how, going from a smaller to a wider gamut working space doesn't lose any information. It just means you're not taking advantage of the wider gamut.

@brecht Here’s a posting from the OCIO listing. Might be useful to add on the proper additional DCC functionality to Blender?

Add OCIO support to FFmpeg or another FOSS project
An important effort for OCIO is expanding the libraries and applications where OCIO is implemented. Users have specifically called out FFmpeg as a library and command-line tool where they would like to see OCIO support, but any other FOSS project leveraged by digital artists or production pipelines could be considered. This could also include upgrading an existing implementation to support the OCIO v2 API and leverage new functionality.
Special skills: Familiarity with prospective host library or tool, and knowledge of working with image buffers and authoring plugins.

https://docs.google.com/document/d/1sa68Cjq0TzSYFg2xhf2w2FFAQ1In7G391iINWKXfDo8/edit

@brecht Here’s a posting from the OCIO listing. Might be useful to add on the proper additional DCC functionality to Blender? >Add OCIO support to FFmpeg or another FOSS project >An important effort for OCIO is expanding the libraries and applications where OCIO is implemented. Users have specifically called out FFmpeg as a library and command-line tool where they would like to see OCIO support, but any other FOSS project leveraged by digital artists or production pipelines could be considered. This could also include upgrading an existing implementation to support the OCIO v2 API and leverage new functionality. >Special skills: Familiarity with prospective host library or tool, and knowledge of working with image buffers and authoring plugins. https://docs.google.com/document/d/1sa68Cjq0TzSYFg2xhf2w2FFAQ1In7G391iINWKXfDo8/edit

Not sure how, going from a smaller to a wider gamut working space doesn't lose any information. It just means you're not taking advantage of the wider gamut.

All of the assumptions for the ACES chain is from AP1 / ACEScg. Going from BT.709 based default primaries to AP1 and then out the door via the RRT etc. doesn’t make any sense here?

Maybe we're talking about different things, but as far as I know we only need our chosen perceptually linear space to be invertible.

Imagine a colour wheel where the image maker wants to select a colour based on greyscale. Here the displayed buffer is not invertible, so the UI buffer would be uni-directional. The mouse pick would peek into the UI buffer pre-rendering transform in this instance.

Same sort of idea would apply for show LUTs etc, where the image maker may want to choose the colour based on how it appears from the show LUT rendering, and those too may not be invertible.

Finally, consider the case of data where the UI would be able to use one of the V2 transforms for data for representation.

> Not sure how, going from a smaller to a wider gamut working space doesn't lose any information. It just means you're not taking advantage of the wider gamut. All of the assumptions for the ACES chain is from AP1 / ACEScg. Going from BT.709 based default primaries to AP1 and then out the door via the RRT etc. doesn’t make any sense here? > Maybe we're talking about different things, but as far as I know we only need our chosen perceptually linear space to be invertible. Imagine a colour wheel where the image maker wants to select a colour based on greyscale. Here the displayed buffer is not invertible, so the UI buffer would be uni-directional. The mouse pick would peek into the UI buffer pre-rendering transform in this instance. Same sort of idea would apply for show LUTs etc, where the image maker may want to choose the colour based on how it appears from the show LUT rendering, and those too may not be invertible. Finally, consider the case of data where the UI would be able to use one of the V2 transforms for data for representation.

Added subscriber: @STEP-ANI-MOTION

Added subscriber: @STEP-ANI-MOTION

Added subscriber: @Aeraglyx

Added subscriber: @Aeraglyx

Added subscriber: @AlexeyAdamitsky

Added subscriber: @AlexeyAdamitsky
Member

Added subscriber: @EAW

Added subscriber: @EAW

The OCIO 2.1 has been released now. Would that be mature enough?
Though, ACES v2's gamut compression is still in a slow progress... and it's planned to be ready in early 2022 according to the official open plan video conf.

The OCIO 2.1 has been released now. Would that be mature enough? Though, ACES v2's gamut compression is still in a slow progress... and it's planned to be ready in early 2022 according to the official open plan video conf.
Author
Owner

Changed status from 'Confirmed' to: 'Archived'

Changed status from 'Confirmed' to: 'Archived'
Author
Owner

Follding into #68926 (Color Management Improvements).

Follding into #68926 (Color Management Improvements).
Contributor

Added subscriber: @Eary

Added subscriber: @Eary
Contributor

A question. Is OCIO v2's aliases feature supported yet? This is an important feature to keep different configs (potentially new official config to ship with Blender) compatible with older *.blend files!
image.png

A question. Is OCIO v2's aliases feature supported yet? This is an important feature to keep different configs (potentially new official config to ship with Blender) compatible with older *.blend files! ![image.png](https://archive.blender.org/developer/F12870315/image.png)

They are.

I've implemented them in the testing trial configuration here...
https://github.com/sobotka/AgX

Seems that "Filmic" as an alias still shits the sheets with:
Color management: scene view "Filmic" not found, setting default "Display Native".

Which I suspect is because it's probably hard coded somewhere? I've included all the aliases.

They are. I've implemented them in the testing trial configuration here... https://github.com/sobotka/AgX Seems that "Filmic" as an alias still shits the sheets with: `Color management: scene view "Filmic" not found, setting default "Display Native".` Which I suspect is because it's probably hard coded somewhere? I've included all the aliases.
Author
Owner

You can define aliases of color spaces, there are no aliases for views as far as I know.

You can define aliases of color spaces, there are no aliases for views as far as I know.

Good point.

In AgX, in order to make it more compatible with other DCCs, I've related all of the transforms into ColorSpaceTransforms. Perhaps it makes sense to point the Blender code at role?

Good point. In AgX, in order to make it more compatible with other DCCs, I've related all of the transforms into `ColorSpaceTransform`s. Perhaps it makes sense to point the Blender code at role?
Contributor

In #84819#1308897, @troy_s wrote:
I've included all the aliases.

I saw it
https://blenderartists.org/t/feedback-development-filmic-baby-step-to-a-v2/1361663/10

For some reason it doesn't seem to work on my end? Or am I mistaken?

In #84819#1308897, @troy_s wrote:
Seems that "Filmic" as an alias still shits the sheets
...
Which I suspect is because it's probably hard coded somewhere?

Would this also be the reason for Non-Color/Generic Data etc. not working on my end?
image.png

> In #84819#1308897, @troy_s wrote: > I've included all the aliases. I saw it https://blenderartists.org/t/feedback-development-filmic-baby-step-to-a-v2/1361663/10 For some reason it doesn't seem to work on my end? Or am I mistaken? > In #84819#1308897, @troy_s wrote: > Seems that "Filmic" as an alias still shits the sheets > ... > Which I suspect is because it's probably hard coded somewhere? Would this also be the reason for Non-Color/Generic Data etc. not working on my end? ![image.png](https://archive.blender.org/developer/F12870329/image.png)

Added subscriber: @Rockbard

Added subscriber: @Rockbard

Added subscriber: @Emi_Martinez

Added subscriber: @Emi_Martinez
Thomas Dinges added this to the 2.90 milestone 2023-02-08 16:25:39 +01:00
Sign in to join this conversation.
No Label
Interest
Alembic
Interest
Animation & Rigging
Interest
Asset Browser
Interest
Asset Browser Project Overview
Interest
Audio
Interest
Automated Testing
Interest
Blender Asset Bundle
Interest
BlendFile
Interest
Collada
Interest
Compatibility
Interest
Compositing
Interest
Core
Interest
Cycles
Interest
Dependency Graph
Interest
Development Management
Interest
EEVEE
Interest
EEVEE & Viewport
Interest
Freestyle
Interest
Geometry Nodes
Interest
Grease Pencil
Interest
ID Management
Interest
Images & Movies
Interest
Import Export
Interest
Line Art
Interest
Masking
Interest
Metal
Interest
Modeling
Interest
Modifiers
Interest
Motion Tracking
Interest
Nodes & Physics
Interest
OpenGL
Interest
Overlay
Interest
Overrides
Interest
Performance
Interest
Physics
Interest
Pipeline, Assets & IO
Interest
Platforms, Builds & Tests
Interest
Python API
Interest
Render & Cycles
Interest
Render Pipeline
Interest
Sculpt, Paint & Texture
Interest
Text Editor
Interest
Translations
Interest
Triaging
Interest
Undo
Interest
USD
Interest
User Interface
Interest
UV Editing
Interest
VFX & Video
Interest
Video Sequencer
Interest
Virtual Reality
Interest
Vulkan
Interest
Wayland
Interest
Workbench
Interest: X11
Legacy
Blender 2.8 Project
Legacy
Milestone 1: Basic, Local Asset Browser
Legacy
OpenGL Error
Meta
Good First Issue
Meta
Papercut
Meta
Retrospective
Meta
Security
Module
Animation & Rigging
Module
Core
Module
Development Management
Module
EEVEE & Viewport
Module
Grease Pencil
Module
Modeling
Module
Nodes & Physics
Module
Pipeline, Assets & IO
Module
Platforms, Builds & Tests
Module
Python API
Module
Render & Cycles
Module
Sculpt, Paint & Texture
Module
Triaging
Module
User Interface
Module
VFX & Video
Platform
FreeBSD
Platform
Linux
Platform
macOS
Platform
Windows
Priority
High
Priority
Low
Priority
Normal
Priority
Unbreak Now!
Status
Archived
Status
Confirmed
Status
Duplicate
Status
Needs Info from Developers
Status
Needs Information from User
Status
Needs Triage
Status
Resolved
Type
Bug
Type
Design
Type
Known Issue
Type
Patch
Type
Report
Type
To Do
No Milestone
No project
No Assignees
16 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#84819
No description provided.