* Do not remove this line (it will not be displayed)
{:toc}
# Introduction
While writing some code to add KTX2 support to OpenImageIO (industry-standard
open-source software for manipulating images in a plethora of formats), I
stumbled across many values a colorspace attribute can be set to but I only
understood (not really, surface level understanding) only `srg_rec709_scene`
(I didn't even understand what the `rec709` is and just assumed that this
refers to sRGB). Looking at OpenImageIO's documentation of `Color information`
didn't help either:
```rst
.. option:: "oiio:ColorSpace" : string
The name of the color space of the color channels. This can be the name of
any documented Color Interop Forum standard token, or any color space,
alias, or role known to OpenColorIO. Common values include:
- `"lin_rec709_scene"`, : Color pixel values are known to be linear
scene-referred and using sRGB/Rec709 color primaries. Note that
`"lin_rec709"` is treated as a synonym.
- `"lin_ap1_scene"`, `"ACEScg"` : ACEScg color space encoding.
- `"lin_ap0_scene"` : ACES2065-1, the recommended ACES space for
interchange and archiving.
- `"srgb_rec709_scene"` : Using standard (piecewise) sRGB response and
primaries. The token `"sRGB"` is treated as a synonym.
- `"g22_rec709_scene"` : Rec709/sRGB primaries, but using a response curve
corresponding to gamma 2.2.
Additionally, `"scene_linear"` is a role that is appropriate for color
pixel values are known to be scene-linear and using facility-default color
primaries as defined by the OpenColorIO configuration.
```
I can't reasonably submit a PR to a major CG repository and expect to only
support sRGB (potentially wrongly so).
Off to Google we go! Let's search `color space`, or `understanding color spaces`,
or `Rec709`, or `sRGB`.
At this point I have to state the following fact: all (yes, all - except maybe
Wikipedia but using that to understand a new concept is usually very rough.
And dull. And boring) articles in at least first 3 pages of
Google/DuckDuckGo/Bing (insert other garbage search engines here) are absolute
garbage LLM slop that regenerate the same very high-level *explanation* points.
Off to books/papers we go. A single search through Reddit (yes, it really is
significantly much better that any search engine what so ever because of the
community. Or at least what remained of it) and I find this goldgem of a paper:
[Cinematic Color][cinematic-color].
Throughout this guide, I will try to complement the paper with areas that I
think are not explained enough to beginners. I am writing this as I read
through it and try to understand it. This approach offers better understanding,
I believe, of areas that were assumed to be already known by the author (nothing
wrong with that, not every single book has to be written for absolute
beginners).
If you notice any errors/mistakes/typos, please see the bottom of the page the
link to the Markdown file in the Github repository from which this site is
generated. Any PRs are more than welcome :-)
## Human Visual System
Light is an electromagnetic spectrum and we (as in humans) are only able to
*perceive* a portion of it; namely 380-780nm. Color is the way our brain
reacts/encodes such information. It could have anything else (maybe perceive
such wavelengths through sound?); doesn't matter. What matters is that color
in reality doesn't exist and only exists through our perception of certain light
wavelengths.
Apparently, in 1931, some researchers tried to answer the following question:
[cinematic-color]: https://raw.githubusercontent.com/jeremyselan/cinematiccolor/master/siggraph2012/cinematic_color.pdf
[cie-1931]: https://en.wikipedia.org/wiki/CIE_1931_color_space
* Do not remove this line (it will not be displayed)
{:toc}
# Introduction
While writing some code to add KTX2 support to OpenImageIO (industry-standard
open-source software for manipulating images in a plethora of formats), I
stumbled across many values a colorspace attribute can be set to but I only
understood (not really, surface level understanding) only `srg_rec709_scene`
(I didn't even understand what the `rec709` is and just assumed that this
refers to sRGB). Looking at OpenImageIO's documentation of `Color information`
didn't help either:
```rst
.. option:: "oiio:ColorSpace" : string
The name of the color space of the color channels. This can be the name of
any documented Color Interop Forum standard token, or any color space,
alias, or role known to OpenColorIO. Common values include:
- `"lin_rec709_scene"`, : Color pixel values are known to be linear
scene-referred and using sRGB/Rec709 color primaries. Note that
`"lin_rec709"` is treated as a synonym.
- `"lin_ap1_scene"`, `"ACEScg"` : ACEScg color space encoding.
- `"lin_ap0_scene"` : ACES2065-1, the recommended ACES space for
interchange and archiving.
- `"srgb_rec709_scene"` : Using standard (piecewise) sRGB response and
primaries. The token `"sRGB"` is treated as a synonym.
- `"g22_rec709_scene"` : Rec709/sRGB primaries, but using a response curve
corresponding to gamma 2.2.
Additionally, `"scene_linear"` is a role that is appropriate for color
pixel values are known to be scene-linear and using facility-default color
primaries as defined by the OpenColorIO configuration.
```
I can't reasonably submit a PR to a major CG repository and expect to only
support sRGB (potentially wrongly so).
Off to Google we go! Let's search `color space`, or `understanding color spaces`,
or `Rec709`, or `sRGB`.
At this point I have to state the following fact: all (yes, all - except maybe
Wikipedia but using that to understand a new concept is usually very rough.
And dull. And boring) articles in at least first 3 pages of
Google/DuckDuckGo/Bing (insert other garbage search engines here) are absolute
garbage LLM slop that regenerate the same very high-level *explanation* points.
Off to books/papers we go. A single search through Reddit (yes, it really is
significantly much better that any search engine what so ever because of the
community. Or at least what remained of it) and I find this goldgem of a paper:
[Cinematic Color][cinematic-color].
Throughout this guide, I will try to complement the paper with areas that I
think are not explained enough to beginners. I am writing this as I read
through it and try to understand it. This approach offers better understanding,
I believe, of areas that were assumed to be already known by the author (nothing
wrong with that, not every single book has to be written for absolute
beginners).
If you notice any errors/mistakes/typos, please see the bottom of the page the
link to the Markdown file in the Github repository from which this site is
generated. Any PRs are more than welcome :-)
## Human Visual System
Light is an electromagnetic spectrum and we (as in humans) are only able to
*perceive* a portion of it; namely 380-780nm. Color is the way our brain
reacts/encodes such information. It could have anything else (maybe perceive
such wavelengths through sound?); doesn't matter. What matters is that color
in reality doesn't exist and only exists through our perception of certain light
wavelengths.
Apparently, in 1931, some researchers tried to answer the following question:
[cinematic-color]: https://raw.githubusercontent.com/jeremyselan/cinematiccolor/master/siggraph2012/cinematic_color.pdf
[cie-1931]: https://en.wikipedia.org/wiki/CIE_1931_color_space