Current Server Time: 17:47 (Central Europe)

#apertus IRC Channel Logs

2015/02/10

Timezone: UTC


23:00
dmjnova
left the channel
23:01
dmjnova
joined the channel
23:05
ItsMeLenny
joined the channel
23:17
dmjnova
left the channel
23:23
mithro
joined the channel
23:52
cfelton
joined the channel
00:07
dmjnova
joined the channel
00:25
fsteinel_
joined the channel
00:28
fsteinel
left the channel
00:59
rhavan_
joined the channel
00:59
rhavan
left the channel
00:59
rhavan_
changed nick to: rhavan
01:52
troy_s
joined the channel
03:14
dmjnova
left the channel
03:38
dmjnova
joined the channel
07:56
niemand
joined the channel
08:20
fsteinel_
changed nick to: fsteinel
08:21
ZeroQ
joined the channel
08:25
ZeroQ
left the channel
08:34
niemand
left the channel
09:39
ItsMeLenny
left the channel
09:54
Bertl
off to bed now ... have a good one everyone!
09:56
Bertl
changed nick to: Bertl_zZ
10:21
ItsMeLenny
joined the channel
10:46
se6astian|away
changed nick to: se6astian
10:55
se6astian
changed nick to: se6astian|away
11:31
niemand
joined the channel
11:55
se6astian|away
changed nick to: se6astian
12:53
niemand
left the channel
14:25
niemand
joined the channel
14:36
niemand
left the channel
14:39
ItsMeLenny
left the channel
16:19
Bertl_zZ
changed nick to: Bertl
16:19
Bertl
morning folks!
16:21
se6astian
hello :D
18:38
aombk
hi
18:43
rhavan
left the channel
18:45
rhavan
joined the channel
18:46
fsteinel
hi
19:03
intracube
evening
19:03
intracube
has anyone used flowblade?
19:24
troy_s
intracube: Nope. Seen it.
19:50
intracube
troy_s: yeah, I've been watching a few video tuts. looks very promising
19:50
intracube
with the inclusion of ripple/roll tools
19:52
intracube
shows they're going for the important things
19:53
intracube
instead of 1980s barn-door wipe effects
19:58
troy_s
intracube: Important things is not fscking around with damn codecs and crap.
19:58
troy_s
intracube: Just cut dammit. Just cut.
19:58
troy_s
intracube: The chat I had in PiTiVi was blood rising.
20:02
intracube
troy_s: ^ I missed that
20:02
troy_s
intracube: It was ugly.
20:02
intracube
but as long as it can read a few key codecs like prores
20:02
intracube
heh
20:03
troy_s
intracube: I suggested a simple damn thing which was A) How is your color blending handled?
20:03
troy_s
And I looked after was linked to the code.
20:03
troy_s
The code in gstreamer (IIRC) basically does a (telltale idiocy when things are labelled YUV :) ) rgb_to_yuv
20:03
troy_s
as in the blends are done on a hard coded rgb to YUV transform.
20:04
troy_s
WHICH, if it were an offline editor, is no big ass deal. It's all proxy crap.
20:04
troy_s
BUT, it aims to be an online thing... so let me see... here's my thought process immediately:
20:04
troy_s
IF it is a hardcoded RGB to YCbCr transform:
20:04
troy_s
1) It is broken. It needs to abide by the source YCbCr obviously.
20:04
troy_s
2) It is broken. It needs to linearize the values to display linear before compositing.
20:05
troy_s
Etc.
20:05
troy_s
To which I was (thankfully) educated on the fact that color is subjective.
20:05
troy_s
No shit.
20:05
intracube
didn't realise it did compositing
20:05
troy_s
The lead developer actually linked me to a video where it describes how color is subjecive.
20:05
troy_s
Compositing == blending
20:05
troy_s
As in dissolves
20:05
troy_s
fades
20:05
troy_s
overs
20:05
troy_s
anything
20:06
troy_s
That's compositing. Not what you expect when you hear "Compositor" per se of course, but the composition of images.
20:06
intracube
oh, isn't gamma encoded more suited to crossfades like that?
20:06
troy_s
Think about that for a second.
20:06
troy_s
It is precisely similar to gradients actually.
20:06
intracube
so they appear 'linear'
20:06
troy_s
On one hand, if our interface uses a linear line
20:06
intracube
thinks we already had this discussion :)
20:06
troy_s
to indicate a crossfade
20:06
troy_s
Yep.
20:06
troy_s
But worth revisiting because it has all sorts of cool little bits in it.
20:06
intracube
eh, but not right now maybe :)
20:07
troy_s
Crossfade from A to B with a linear line UI.
20:07
troy_s
Middle might be assumed to be about 50% perceptual intensity. But the colors don't mix correctly unless they are display linearized. Enter problem. :)
20:08
intracube
troy_s: https://www.youtube.com/watch?v=oSgnnkraMIg
20:08
intracube
gamma encoded text fade on the left, true linear on the right
20:08
intracube
actually, the right might be Log
20:09
intracube
pretty sure it is...
20:09
troy_s
The solution isn't quite as easy to see initially: A) Take the spline, and IF it is assumed to be representing a perceptual intensity spline, run that through the inverse transfer curve to bend the spline values to linearized format. B) Blend on linear reference color values according to that value. :) )
20:09
intracube
but same difference I think
20:09
troy_s
Their log is screwed
20:09
troy_s
Log has a white value generally
20:09
intracube
"Take the spline, and IF it is assumed to be representing a perceptual intensity spline, run that through the inverse transfer curve"
20:09
troy_s
their upper limit didn't get reached to it.
20:09
intracube
yeah, you mentioned that last time :)
20:09
troy_s
I suspect they took an sRGB image and rolled it to a log2 or something? Leaving too much headroom.
20:09
intracube
still adds complexity - user might want to remain oblivious to that level of whats going on
20:10
intracube
'behind the scenes'
20:10
troy_s
Well it depends.
20:10
troy_s
I don't ever see a user. I see an artist or experienced peep generally.
20:10
troy_s
And the bottom line is that you can sort of "hide" that.
20:11
troy_s
Assumption A: That the artist expects a bargraph / UI element to be a perceptual human oriented thing. B: That the artist expects correct blending.
20:11
intracube
still finds blender's handling of linear gradients frustrating
20:11
troy_s
Gradients are deadly tricky in the same way
20:11
intracube
but this is #apertus - not going to complain about that here
20:11
intracube
:)
20:11
intracube
and food is imminent
20:11
troy_s
It's all relevant really. Color and imaging and such.
20:11
troy_s
Gradients have to track in linear color for the color traversal
20:11
troy_s
Be it opponent or tri-color.
20:11
troy_s
But the intensity needs to be likely perceptual in many instances
20:12
troy_s
So it is the same idea - take the values and run them inversely through whatever perceptual / ldr intensity transfer is used, then march through the linearized colors.
20:12
troy_s
Often implementations only do HALF of that, or worse, neither.
20:15
intracube
yep
20:15
intracube
would be good if Blender could have a gamma transform toggle for the gradient texture
20:15
intracube
for example
20:20
troy_s
intracube: The problem is that I guess folks get very confused with color very quickly. Or rather, these sorts of things are an immediate and visceral "What you thought you knew about color was totally wrong" feeling.
20:21
troy_s
intracube: I'm just about to begin my second pass on the color wheels, which have, shocker, the same issue.
20:22
troy_s
intracube: My brain hurts though. I'm unsure exactly the correct approach.
20:22
troy_s
I want to take the circular part with the colors laid out (which are effectively linearized reference values) roll them through the inverse of the artist-selected transfer curve (so they are correctly nonlinearized)
20:23
troy_s
then roll them through the output transform. But then you can have a mismatch. Example - roll them through a log inverse, then they go to sRGB is... erf. Not sure it really is an issue is the issue. :)
20:24
intracube
head explodes
20:24
intracube
do you mean color picker or compositor 3 point color correction?
20:30
intracube
troy_s: but good luck battling the code
20:30
intracube
and sergey :P
20:32
troy_s
I mean the HSV wheel.
20:32
troy_s
You know the picker?
20:32
troy_s
The round thing with the value slider?
20:32
troy_s
Same sort of issues where the two halves of color collide into a complex little cocktail.
20:33
troy_s
intracube: An HSV wheel needs to be color corrected. For screen (display corrections) and for general colorspace transforms (from a different gamut to the output gamut)
20:35
troy_s
intracube: So that means you have to roll it through the transforms. The problem is, if you lay it out in display referred values, 0..1, you are already assuming it is in a suitable for display format, aka nonlinear. If you roll that through the output transforms, well you get a doubleup and that's ghastly.
20:37
intracube
continue in #blendervse ?
21:40
se6astian
time for bed
21:40
se6astian
good night
21:40
se6astian
changed nick to: se6astian|away