[Sidefx-houdini-list] OT: sRGB to linear question

Edward Lam edward at sidefx.com
Tue Jul 8 10:38:46 EDT 2008


sRGB is NOT any worst or less accurate. It's what you might do with the 
non-linear image that *might* be a problem. For a striking example of 
what can go wrong if you perform image algorithms that assume linear 
space with non-linear images, see this link:
    http://www.4p8.com/eric.brasseur/gamma.html [1]

As usual, it's only a problem if you don't know what you're doing. :)

Cheers,
-Edward

1. Note that (AFAICT) the author of the link doesn't try properly gamma 
calibrate the software he tests.

John Coldrick wrote:
> On Tuesday 08 July 2008 09:51, Sean Lewkiw wrote:
>> A question then... let's say I want an object that is a nice purply blue
>> at the top and a greenish orange at the bottom.  In one render I achieve
>> this using point colours.  In another I achieve exactly the same thing
>> using an sRGB image I downloaded off the interweb.  How is the one
>> derived from the sRGB image any worse or less accurate than the one
>> derived from the point colours?  Both final renders are linear.
> 
> 	If they are the same colour, then they're the same.  sRGB doesn't inherently 
> taint anything - if it looks correct, and it goes out the final pipeline 
> looking correct, then it's no more right or wrong than the point colour 
> version.
> 
> 	It seems obvious, but where it gets tricky is that when you render that sRGB 
> texture in mantra, what you're doing is rendering a texture map that has a 
> gamma burned into it(sRGB) that is different than what the renderer is 
> using(probably linear assuming you've changed nothing).  Now, you don't see 
> any problem because the monitor you're screening mplay on *is* sRGB(most 
> likely, unless you're using a calibration program).  If you're sending that 
> image, like we frequently do here, to an inferno suite that's also working 
> with tiff plates that have sRGB burned into it, you can merrily carry on and 
> not worry about things.  Because mantra is linear by default, it's 
> essentially passing things through it without changing anything.
> 
> 	Strictly speaking, though, this is 'wrong'(and is where you get much finger 
> wagging and tsk-tsk'ing).  You should be converting the tmap to linear, and 
> rendering it, then viewing it through a calibration program that expects the 
> incoming imagery to be linear, and will convert it to sRGB or whatever gamma 
> your display device is using(or you burn it back in for final should that be 
> the pipeline).  Again, if you're all video all the time, you never do film, 
> you never get sources from cineon format, etc...then this all seems like a 
> lot of worry over nothing, and in fact it is, as long as your pipeline works 
> like that.  It's more work to convert it, render it, comp it, then re-convert 
> it again.  You can probably safely do without it(although some people will 
> still finger wag you regarding the colourspace you're compositing in).  The 
> instant you add other variables to the pipeline, though, you need to be aware 
> of it.
> 
>> I'm not trying to be provocative, just trying to get an understanding of
>> this whole thing.
> 
> 	Believe me, I hear you.  :)  It's a complicated topic and non-obvious, with 
> lots of misinformation out there.
> 
> 	Andy got to that link before me, it's very useful.  Hopefully this post is 
> somewhat complentary as a 'why' explanation as opposed to 'how'.  I found 
> the 'why's to be poorly explained much of the time, based on a film 
> assumption.
> 
> 	Cheers,
> 
> 	J.C.
> 




More information about the Sidefx-houdini-list mailing list