Author Topic: half float 16 bit or full float 32 bits EXR or tiff ?  (Read 19292 times)

2014-05-23, 13:41:35

Fibonacci

  • Active Users
  • **
  • Posts: 271
  • 3Dmanufaktura
    • View Profile
Hello,

any experience ? Which one is the better choice ? For CC and for faster post...or in most cases...

Cheers!
Holy Corona : the materials is the clue.

2014-05-23, 14:18:01
Reply #1

johan belmans

  • Primary Certified Instructor
  • Active Users
  • ***
  • Posts: 719
    • View Profile
    • belly.be
HI

I make ue of EXR - 16 bit in After Effects. the full float option doesn't benefit more for me.


2014-05-24, 15:07:53
Reply #2

Captain Obvious

  • Active Users
  • **
  • Posts: 168
    • View Profile
OpenEXR is better than TIFF, and 16-bit mode is sufficient for all reasonable use cases. Keep in mind though that most software that loads floating point images will load them as 32-bit regardless of the file's bit depth. A 16-bit OpenEXR file will load as a 32-bit file in After Effects, Nuke, Photoshop, etc. The only thing you're saving is hard drive space. But because 16-bit files have more than enough colour precision anyway, you might as well use them and save some space. They probably load faster too.

2014-05-24, 19:11:27
Reply #3

Ondra

  • Administrator
  • Active Users
  • *****
  • Posts: 8864
  • Turning coffee to features since 2009
    • View Profile
There was one rather big problem I encountered, that 16bit EXR cannot save direct sundisk or its specular reflection. Because of that corona saves everything in 32bit
Rendering is magic.
Private scene uploader | How to get minidumps for crashed/frozen 3ds Max | Sorry for short replies, brief responses = more time to develop Corona ;)

2014-05-24, 19:44:27
Reply #4

Fibonacci

  • Active Users
  • **
  • Posts: 271
  • 3Dmanufaktura
    • View Profile
So, what do you offer to the output ?

The space is dosn't matter now, but if the Corona saves everthing in 32 bit...I have made already lot of renders for post, but those has in 16 bit tiff...and if the Corona saves in 32...that's mean I "lost" the half color information?.
It's ok now, because the render time was more than a week...but next time, I just would like to know, what do you recomend.

If I want to make 4K pictures...for the new 4K monitors ( i don't have yet), then the great way sould be the 32 EXR...isn't it?
Holy Corona : the materials is the clue.

2014-05-24, 20:09:11
Reply #5

DeadClown

  • Global Moderator
  • Active Users
  • ****
  • Posts: 1437
    • View Profile
    • racoon-artworks
There was one rather big problem I encountered, that 16bit EXR cannot save direct sundisk or its specular reflection. Because of that corona saves everything in 32bit

What type of saving are you talking about? I made a quick test with max's common render output and got 16 and 32 bit as expected (32 bit exrs are about 4 times the size of a 16 bit exr) so I guess you're referring to the autosave/resume exrs corona is saving?
Any sufficiently advanced bug is indistinguishable from a feature.

2014-05-24, 20:31:49
Reply #6

Ondra

  • Administrator
  • Active Users
  • *****
  • Posts: 8864
  • Turning coffee to features since 2009
    • View Profile
So, what do you offer to the output ?

The space is dosn't matter now, but if the Corona saves everthing in 32 bit...I have made already lot of renders for post, but those has in 16 bit tiff...and if the Corona saves in 32...that's mean I "lost" the half color information?.
It's ok now, because the render time was more than a week...but next time, I just would like to know, what do you recomend.

If I want to make 4K pictures...for the new 4K monitors ( i don't have yet), then the great way sould be the 32 EXR...isn't it?

Actually it is not half the color resolution, the 32bit float resolution is 4096 times finer than 16 bit. But 16 bit is usually enough.

The problem is that 16bit is limited in the terms of maximal possible luminance - standard sun color is brighter than any finite value expressable as 16bit float.


DeadClown: I mean Corona autosave, save/resume/load features, and DR
Rendering is magic.
Private scene uploader | How to get minidumps for crashed/frozen 3ds Max | Sorry for short replies, brief responses = more time to develop Corona ;)

2014-05-25, 00:36:26
Reply #7

Captain Obvious

  • Active Users
  • **
  • Posts: 168
    • View Profile
There was one rather big problem I encountered, that 16bit EXR cannot save direct sundisk or its specular reflection. Because of that corona saves everything in 32bit
The max value is 65536, so anything that's brighter than that will be clamped. I guess if you render with a bit of highlight compression, you'll be able to squeeze in a greater range. It's pretty trivial to remove a known global tonemapper anyway.

2014-05-25, 12:13:22
Reply #8

Ondra

  • Administrator
  • Active Users
  • *****
  • Posts: 8864
  • Turning coffee to features since 2009
    • View Profile
it actually won't get clamped on its own, but turned to INF, which is hella fun to deal with. But maybe 3dsmax is intelligent enough to check and clamp when saving it (hahaahaha)
Rendering is magic.
Private scene uploader | How to get minidumps for crashed/frozen 3ds Max | Sorry for short replies, brief responses = more time to develop Corona ;)

2014-05-25, 12:45:23
Reply #9

DeadClown

  • Global Moderator
  • Active Users
  • ****
  • Posts: 1437
    • View Profile
    • racoon-artworks
But maybe 3dsmax is intelligent enough to check and clamp when saving it (hahaahaha)

Nope ;) You can easily run into this problem if you're rendering a huge (dimensions) scene and try to use the world position pass in post. Exrs are actually saving #inf values.
Any sufficiently advanced bug is indistinguishable from a feature.

2014-05-25, 14:23:05
Reply #10

Captain Obvious

  • Active Users
  • **
  • Posts: 168
    • View Profile
it actually won't get clamped on its own, but turned to INF, which is hella fun to deal with. But maybe 3dsmax is intelligent enough to check and clamp when saving it (hahaahaha)
Yeah that'll happen ;-)

2014-05-25, 14:47:12
Reply #11

blank...

  • Active Users
  • **
  • Posts: 207
    • View Profile
I would say 16 bit half float. You would need some really specific situation to absolutely NEED 32 bit.
Disk space isn't a problem these days, but it's not about disk space. It's about having to work with 30mb file vs. 3mb file. When you start your post processing and get dozens of nodes in there (even hundred or more for the big boys who do really complex stuff), things can get a bit laggy with large files (you will be adding other passes as well).

Corona saves everything as 32bit? Not a problem, in Fusion you can bring down bit depth in the loader it self or in a separate node. There is probably something similar in other compositors.

2014-05-26, 17:49:13
Reply #12

Fibonacci

  • Active Users
  • **
  • Posts: 271
  • 3Dmanufaktura
    • View Profile
I agreed.

Weel...Maybe the 32 bit is unnecessary for the color pass, just 16 enought. But for the transiton beetwen white and black much smoother in ZDepth and AO pass (Guassian blur is ok for that, but...maybe not the same). And I think this is sould be important for the high quality post production.

And 32 same good for the reflections and GI channels...I thinks.
Ofcourse I don't wanna make hundred and hundreds node in a "simple" post. But when I see some really high-end post production...than I feel I have to try to achieve that. Ofcourse, those guys has a lot of experience and practice...

Not mentioned the HDR.
Holy Corona : the materials is the clue.

2015-02-10, 15:11:40
Reply #13

SairesArt

  • Active Users
  • **
  • Posts: 657
  • Pizza | The Cheesen One
    • View Profile
    • SairesArt Portfolio
Weel...Maybe the 32 bit is unnecessary for the color pass, just 16 enought.
Old threads need love too.
I really wanna correct this: If you plan on doing Depth of Field or Motionblur in post, it's essential to have Superwhites, which only work in 32-bit, otherwise the Bokeh from DoF and Lightstreaks from MoBlur will have the wrong color and the wrong brightness.

If you don't Plan on doing DoF or MoBlur in Post (eg. 16-bit exrs), there is no reason to go EXR in the first places, due to compatibilty issues, that may arise. PNG deliver the same color info with less space wasted and a saved color profile. And combining multiple passes in one EXR often screws with Aftereffects. If you are on Nuke, exr's are preffered. If you only Photoshop or AE, just stick with dem bloddy PNG's.

2015-02-10, 16:06:31
Reply #14

DeadClown

  • Global Moderator
  • Active Users
  • ****
  • Posts: 1437
    • View Profile
    • racoon-artworks
I really wanna correct this: If you plan on doing Depth of Field or Motionblur in post, it's essential to have Superwhites, which only work in 32-bit, otherwise the Bokeh from DoF and Lightstreaks from MoBlur will have the wrong color and the wrong brightness.

I (also) really wanna correct this...
16 bit float exrs ("half float") save values between 65504.0 and  5.96^–8  (http://http.developer.nvidia.com/GPUGems/gpugems_ch26.html) . So "superwhites" wise, this is more than you'll probably ever need. The important thing is precision here. For nearly everything 16 bit float (not integer) is enough. color, lighting, gi, reflections doesn't matter, it's precise and fast enough for compositing tasks. The only real exception here are world position and zdepth passes  - since those can easily go beyond those values AND need to be very precise. You may not see it right away but if you use a wpp to displace the rendering you see the precision difference between 16 and 32 bit float.

Quote
If you don't Plan on doing DoF or MoBlur in Post (eg. 16-bit exrs), there is no reason to go EXR in the first places, due to compatibilty issues, that may arise. PNG deliver the same color info with less space wasted and a saved color profile. And combining multiple passes in one EXR often screws with Aftereffects. If you are on Nuke, exr's are preffered. If you only Photoshop or AE, just stick with dem bloddy PNG's.
PNG does not deliver the same information. Png ist not linear (so if you want to do any image operation on it you will have to linearize your image first and that means you have already lost a good amount of information). PNG also does not save anything above 1.0, so as soon as you have some overbright pixels you're f*****. I also don't see embedded color profiles as an advantage, but this is another story.
I'm not using multichannel exrs with AE, so I can't say anthing to that. Photoshop is different - since ps is neither able to work linear nor respect gamma of alphas or able to do predivision you're lost anyways, no matter which format you're using...
« Last Edit: 2015-02-10, 18:08:29 by DeadClown »
Any sufficiently advanced bug is indistinguishable from a feature.

2015-02-10, 16:19:29
Reply #15

Juraj Talcik

  • Active Users
  • **
  • Posts: 3435
  • Tinkering away
    • View Profile
    • studio website
Photoshop is different - since ps is neither able to work linear nor respect gamma of alphas or able to do predivision you're lost anyways, no matter which format you're using...

A big plus for Deadclown to plow through this bro-science :- )

I don't even have energy anymore to explain the difference of linear files...

But regarding Photoshop not working linearly or respect gamma, I would want to say that's also incorrect. While the options for working linearly in PS are still limited, it's more than enough for basic compositing and color correction.
PS opens linear files like .exr completely fines, switches to 32bit/Linear mode and the layer modes work in linear math (those that can). What do you find problematic ?

Since CC was introduced, there were great improvements in this regards, like layers, alphas,etc.. It's not perfect, but it works.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik    My Behance portfolio
lysfaere.com   Something new

2015-02-10, 16:53:32
Reply #16

DeadClown

  • Global Moderator
  • Active Users
  • ****
  • Posts: 1437
    • View Profile
    • racoon-artworks
mhhhh... looks like Photoshop is indeed loading the alpha linear now, so I don't seem to be informed very well. Tested an exr here and it seems fine.  (Edit: I did not yet test non-linear image formats :P)
An example of what I find problematic is that you can't do basic pre-divide / post-multiply actions (I'm not talking about remove black matte), so I'm extremely limited in what I can do compositing-wise. Sure, I can stack up all essential corona REs but that's pretty much it - If I have to combine it with something else, this is where it stops. Even color corrections are kind of limited since I have to unpremultiply my passes before I do any color corrections. I'd expect that the curve and levels etc tools all have their internal predivide/postmultiply routine (which I doubt...) but as soon as I do it myself with layer blending or a brush or anything I NEED to unpremultiply first.

But I must admit that photoshop is a topic which drives me crazy anyways (not only because of lwf)... I'm maybe not the right person to discuss this matter in an objective way :D
« Last Edit: 2015-02-10, 16:58:44 by DeadClown »
Any sufficiently advanced bug is indistinguishable from a feature.

2015-02-10, 17:05:13
Reply #17

Juraj Talcik

  • Active Users
  • **
  • Posts: 3435
  • Tinkering away
    • View Profile
    • studio website
No you're right, it's very limited in this regard, but for us mortals/simpletons :- D simply toying around with still images, it does the job even when it comes to linear formats :- ) Esp. CC made this far more pleasant experience compared to CS6, so looks like they're listening to people outside of pure 2D graphics/photography fields. It's strange they're no able to merge the toolset capability from AE to PS, that would be perfect.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik    My Behance portfolio
lysfaere.com   Something new

2019-02-13, 23:12:57
Reply #18

VisualBoy

  • Users
  • *
  • Posts: 3
    • View Profile
mhhhh... looks like Photoshop is indeed loading the alpha linear now, so I don't seem to be informed very well. Tested an exr here and it seems fine.  (Edit: I did not yet test non-linear image formats :P)
An example of what I find problematic is that you can't do basic pre-divide / post-multiply actions (I'm not talking about remove black matte), so I'm extremely limited in what I can do compositing-wise. Sure, I can stack up all essential corona REs but that's pretty much it - If I have to combine it with something else, this is where it stops. Even color corrections are kind of limited since I have to unpremultiply my passes before I do any color corrections. I'd expect that the curve and levels etc tools all have their internal predivide/postmultiply routine (which I doubt...) but as soon as I do it myself with layer blending or a brush or anything I NEED to unpremultiply first.

But I must admit that photoshop is a topic which drives me crazy anyways (not only because of lwf)... I'm maybe not the right person to discuss this matter in an objective way :D

And has something changed in your approach after the PS 2018/19 exit? Do you still think that when it comes to uncompromising high-end post-processing with Corona RE, it's just AE or Nuke? :) I've been sitting in the graphics for about 20 years and thoughts with a difference in the initial image for clients are really invisible :) Postprocessing in AE or Nuke is associated with 2-3x longer process than in PS

2019-03-05, 10:48:55
Reply #19

RomasNoreika

  • Active Users
  • **
  • Posts: 136
    • View Profile
    • romasnoreika
Open EXR 16bit more than enough. Works every time.

32bit only if you need that and the software will support it.


But 16bit you will not have any problems with masking, speed, size and so on.

Been using this for a lot of years. :)

32bit is sometimes usefull for people who dont make the lighting work 100% in max (raw render) and then you can really play with different exposures in photoshop and collapse it to 16 bit. With 32 bit you will have a lot of info in the image you can go crazy with it.