Advertisement
Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- 03:57 <kaito> well, i am doing venom's work now. he should talk to you and give you example file
- 03:58 <ideasman42> just try comment the line that does CM
- 03:58 <kaito> i just communicate
- 03:58 <kaito> and share that sybren found a difference in CM for both options
- 03:59 <ideasman42> ok, change here is really 10min work once we know whats the problem exactly
- 03:59 <ideasman42> or less... simple stuff, just agree on proper behavior and implement
- 03:59 <kaito> yeah :) will ask venom to contact sergey monday
- 04:00 <ideasman42> making it same as viewport is reasonable
- 04:01 <ideasman42> was going to make viewport dof possible from sequencer too
- 04:02 <ideasman42> but would like if artists communicate a bit about whats needed/useful/notworking
- 04:03 <ideasman42> they can report bugs on this stuff too
- 04:04 → matthe, Donnerland, meta-androcto, DarkDefender, vekoon, Slasher006pi and howlymowly joined ⇐ light_, mib2berlin_ghost, abergmeier and lukas_t quit
- 04:24 <hackerman-> ideasman42: sequencer works in the display space
- 04:25 <hackerman-> ideasman42: which makes it quite simple to visualize buffers from the sequener in the preview
- 04:25 <hackerman-> ideasman42: however, if you're saving, ay, exr files, you needto perform lineratization
- 04:25 <ideasman42> yep, probably for PNG we can have a fastpath though
- 04:25 ⇐ DingTo quit (~Thomas@dslb-094-218-230-152.094.218.pools.vodafone-ip.de) Read error: Connection reset by peer
- 04:25 <hackerman-> yes, for png it should all be fast
- 04:25 <ideasman42> currently it gets float buffer from imbuf, when it doesnt need to
- 04:26 <ideasman42> (not necessarily anyway)
- 04:26 <hackerman-> float buffer in sequencer is also non-linear
- 04:26 → abergmeier joined (~andreas@p4FE5D13A.dip0.t-ipconnect.de)
- 04:26 <hackerman-> so coversion for save should be fast
- 04:27 <hackerman-> ideasman42: or kinda depends on definition of "gets float buffer"
- 04:27 <hackerman-> OCIO API only operates with float buffers. so _if_ one need to perform conversion, float buffer is to be created
- 04:28 <hackerman-> it's not full buffer for the conversion tho, it's like 64 scanlines per thread
- 04:28 <hackerman-> there are also shortcuts to avoid any computation whrn converting color space to the same one
- 04:28 <hackerman-> so sRGB->sRGB converison will do nothing
- 04:29 <hackerman-> anyway, didn't read full conversation. but feel free to report a bug ans assign to me ;)
- 04:29 <ideasman42> yep, this should really be bug report
- 04:29 <ideasman42> not hearsay second hand info
- 04:29 → KAHR-Alpha joined (~Alpha@AReims-652-1-195-32.w86-192.abo.wanadoo.fr)
- 04:30 <ideasman42> kaito, kick artists to report issues more
- 04:30 <ideasman42> hackerman-, checking further... if we really wanted we could make it faster, and bypass the RenderResult
- 04:31 <psy-fi> I might play the devil's advocate here and say that for correct blending sequencer probably needs to have a linear working space
- 04:31 <ideasman42> avoid float conversion entirely
- 04:31 <psy-fi> CM should probably only be applied at the end
- 04:31 <psy-fi> not during storage of caches
- 04:32 <hackerman-> that's not really true
- 04:33 ⇐ howlymowly quit (~tom@x4d0cb72c.dyn.telefonica.de) Ping timeout: 250 seconds
- 04:33 <psy-fi> meh, if it wasn't for storage and precision issues I would be all for it
- 04:34 <ideasman42> kaito, http://download.blender.org/ftp/ideasman42/donelist/2015.html#week-273-november-9
- 04:34 — psy-fi is curious how fast 16bit EXR operations might be, in fact
- 04:34 <hackerman-> eh, donelist..
- 04:34 → jpbouza joined (~jpbouza@190.172.17.158)
- 04:34 <hackerman-> psy-fi: nice SIMDed half-float is fast
- 04:35 <psy-fi> hackerman-, might be worth considering then?
- 04:35 <hackerman-> totally
- 04:35 <hackerman-> in compo and sequencer
- 04:37 → panzergame and lukasstockner97 joined ⇐ abergmeier quit
- 04:42 <hackerman-> kaito: http://wiki.blender.org/index.php/User:Nazg-gul/Foundation/2015#Week_205:_9th_-_15th_November
- 04:43 → trigrou joined ⇐ Aidy_Burrows, meta-androcto and boud quit
- 04:48 <kaito> psy-fi: people who edit videos expect the edits and operations to work in display space
- 04:48 <kaito> grading, same
- 04:48 <psy-fi> kaito, what operations are those?
- 04:48 <kaito> linear space has been invented for working with additive and subtractive things (combining layers of renders)
- 04:49 <kaito> it's like paint programs
- 04:49 <psy-fi> Also footage and masks>
- 04:49 <kaito> you work on display color there, not linear
- 04:50 <kaito> compositing in linear space is already confusing enough, and the biggest mistake is color correcting in linear
- 04:50 <kaito> color correcting and grading is for display spaces
- 04:50 — hackerman- rememebrs the bug hapened in the VSE for couple of hours, making Ian to grade in linear...
- 04:51 <psy-fi> Depends on what you want to do with sequencer of course
- 04:51 <kaito> edit rendered shots into films
- 04:51 <kaito> that
- 04:51 <psy-fi> kaito, sure, you pick colors in display space, but operations happen in linear
- 04:51 <kaito> or storyboards
- 04:51 <kaito> or get your wedding vid and make a quick edit :)
- 04:52 <kaito> psy-fi: not really
- 04:52 <psy-fi> OK, I see your point, some operations are done on final color
- 04:52 <hackerman-> well, majority of them, actually
- 04:53 <hackerman-> blending is defined in both linear and final spaces
- 04:53 <kaito> color theory is huge headache stuff
- 04:53 <psy-fi> it is
- 04:53 <kaito> unfortunately render code is also affected now ;)
- 04:54 <kaito> i bet vulkan has CM
- 04:54 <hackerman-> kaito: ??
- 04:54 <psy-fi> well, shaders is just numbers
- 04:54 <psy-fi> GL has some sRGB texture formats though
- 04:54 <psy-fi> that supposedly take care of correct conversion
- 04:54 <hackerman-> kaito: even GLSL "has" CM, it's just a LUT lookup function, few lines of code ;)
- 04:55 <psy-fi> but even then there's some freedom on whether texture filtering happens before or after CM
- 04:56 → smellslikedonkey and mib2berlin joined
- 04:56 <kaito> hackerman-: our code has, but the rgb definition of opengl is undefined?
- 04:56 <kaito> i dont think they did anything keeping display display and render space sparated
- 04:57 <hackerman-> just numbers
- 04:57 <kaito> yep
- 04:57 <psy-fi> it's the operations you do that assume certain spaces
- 04:57 <psy-fi> so if you'redoing blending or shading, you'd better assume you're in linear space
- 04:57 <psy-fi> all this did not matter so much in past btw
- 04:57 <hackerman-> not sure why you expect any semantic from vulkan actually
- 04:57 <psy-fi> but now with PBR it becomes more relevant
- 04:57 <hackerman-> they were getting rid of this instead to avoid overhead
- 04:58 → abergmeier joined ⇐ marcclintdion quit
- 04:58 <psy-fi> easily. It's like nodes, throw responsibility to user/shader writer ;)
- 04:59 <psy-fi> "OpenGL has no CM!" "I don't know man, it's your shader, add it in if you want it"
- 05:00 ⇐ t-ask quit (~t-ask@188.165.85.239) Ping timeout: 240 seconds
- 05:00 <kaito> yep if you want PBR in viewports, color (at least linear color) becomes relevant
- 05:00 <kaito> but then things get very muddy
- 05:01 <kaito> i would have thought the vulkan designers would facilitate it in a way you dont have to worry much
- 05:01 → meta-androcto joined (~chatzilla@1.136.96.155)
- 05:01 <kaito> and in a way you can kick out ocio :)
- 05:03 <psy-fi> From what I hear they would force graphics programmers to write in assembly if they could :p
- 05:03 <psy-fi> kidding, it's not that bad
- 05:06 <hackerman-> kaito: not really
- 05:07 ⇐ max12345 quit (~max@x5ce10078.dyn.telefonica.de) Quit: Leaving for now...
- 05:07 <kaito> well, i agree, vulkan is too low level
- 05:07 <hackerman-> kaito: one of major goals about vulkan -- performance gain by avoid driver-side state machine and "context" checks
- 05:08 <kaito> i bet you can make a sane and fast color pipeline then
- 05:08 <hackerman-> yes, like in rendering
- 05:09 <hackerman-> pre-process textures when needed, do shaders in whatever-working-space you consider for self, apply CM on the result if needed
- 05:10 <hackerman-> how more sane and fast it can be? :)
- 05:10 ⇐ meta-androcto quit (~chatzilla@1.136.96.155) Read error: Connection reset by peer
- 05:10 <psy-fi> ha!
- 05:11 → meta-androcto and Guest49883 (was max12345) joined ⇐ danni, Bollebib and SaphireS quit
- 05:15 <psy-fi> so I guess it makes sense to keep sequencer in display space for now
- 05:15 ⇐ Idur2 quit (~idur@ipbcc3fe89.dynamic.kabel-deutschland.de) Ping timeout: 240 seconds
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement