[Bf-funboard] New compositing and VSE monitor tools

[Troy Sobotka]

On Fri, Feb 19, 2016, 2:45 AM David McSween <3pointedit at gmail.com> wrote:

> The color space issues are non-trivial
> in the VSE as I'm told that it must perform its transforms as quickly as
> possible to achieve real time playback.

[...]

> Troy's language may seem opaque
>

Hard to direct discussion when the audience's experience and knowledge is
so diverse.

Here is the easiest way to understand what is meant by “Must do
manipulations on a linearized reference.” Nothing speaks louder to imagers
than imagery, so...

If everyone reading this, completely grasps the problem by the end of this
mail, we can hopefully build on the more nuanced problems.

1) Make a five pointed red star against a fully transparent background. The
star should be pure, fully display referred intense sRGB red. That is, red
at display referred 1.0 or 255 depending on your application. Use a mask or
however you like to make the star.

2) Blur the star in the compositor. We can save it as an EXR if desired.
Blur it just slightly enough to fuzz the points of the star.

3) In the VSE, make a colour strip that is full display referred intensity
cyan. That is, set sRGB blue and sRGB green to full intensity 1.0.

4) Now load the star and perform an alpha-over blend with the star in the
foreground in the VSE.

Observe. See the horrible dark fringing in the blurred region of the star?
This has nothing to do with alpha format or such, and everything to do with
not operating on a linearized reference.

All manipulations in the VSE, from overs to dissolves to you name it, are
broken in this way.

If one can understand and see the brokenness in this simple example, the
answer to “Why a linearized reference model?” is now clear; the colours are
not mixing as they do “correctly” in reality.

With respect,
TJS

>