1. Topics
  2. Question about IIF config

Question about IIF config

Jeremy Selan <jeremy...@...>
 

Marie,

Thanks for catching this! The currently shipping IIF configuration
is based on an older specification (I forget the exact version, but
it's around 8 months old), so my hope is the versions will explain the
differences you are seeing. We've wanted to update to the latest RRT
for awhile now, so let us take a stab at it and we'll see if it fixes
everything. We will also take care to note in the profile which RRT
version it is specifically, to help avoid ambiguity in the future.

The noise (discontinuity) you are seeing in the 3d RRT lut is an
artifact of how we originally generated this table, we'll make sure
it's all fixed in the updated version.

The AllocationTransform you mention describes a linear -> log
transform (a perfect mathematical log operator), where the range from
( 2^-10.4739 , 2^5.52607) is rolled into (0.0-1.0) for sampling with
a 3d lut. The lut was generated by taking a 3d lattice image (0-1) ,
unwarping it through the inverse of the transform, and then applying
the analytical rrt.

Thanks!

-- Jeremy


2011/10/4 Marie Fétiveau <m...@...>:

Hello all !
I'm having a look at IIF config.ocio and I was wondering how you build
the aceslg2_to_Rec709.cube LUT for the rrt_odt_r709 colorspace.
I'm asking because when I compare an OCIOColorspace node (in : aces, out :
rrt_odt_r709) and two TuttleCTL nodes (one with the RRT 2.2.1 and another
with the REC701 softproof ODT), there are some differences :
- a small shift
- + some noise in yellows with OCIOColorspace node.
Here's a screenshot of my nuke scene and a jpg export with TuttleCTL nodes
and an OpenColorIO Node.
For my tests, I'm using an XRite colorchart shot with a Red One and exported
in EXR thanks to REDCineX (ACES option enabled).
am I mis-using the OCIOColorspace node ?
Anyway, I will be pleased to understand how the LUT was processed and
configured in the ocio config file.
I don't fully understand the purpose of the AllocationTransform vars before
the FileTransform :
- !<ColorSpace>
    name: rrt_odt_r709
    family: rrt_odt_r709
    bitdepth: 32f
    isdata: false
    allocation: uniform
    allocationvars: [0, 1]
    from_reference: !<GroupTransform>
      children:
        - !<AllocationTransform> {allocation: lg2, vars: [-10.4739,
5.52607]}
        - !<FileTransform> {src: aceslg2_to_Rec709.cube, interpolation:
linear}


Thanks a lot !
Marie
Marie Fétiveau <m...@...>
 

Thanks for the quick answer and the explanation :)

I tried this afternoon to create my own aceslg2_to_Rec709 LUT but I'm still not sure to understand how it should work.

I started to give a quick try with the Nuke Log2Lin node to see what happened.

So I connected a CMSTestPattern + a Nuke LogToLin node + my 2  CTL node (RRT + ODT) + a generate LUT to create a logToLin_RRT_ODTr709.cube.
I also create a linToLog.cube (which correspond to Nuke LinToLog).

And I set the rrt_odt_r709 colorspace like this : 
from_reference: !<GroupTransform>
      children:
        - !<FileTransform> {src: linToLog.cube, interpolation: linear}
        - !<FileTransform> {src: logToLin_RRT_ODTr709.cube, interpolation: linear}  

When I compare the two CTL nodes and the OCIOColorspace, that's not that bad but there's a kind of offset + gamma shift...
 A picture is often better than a long talk away (I put the viewer in sRGB instead linear to emphasize the shift on the screenshot).

Where's my mistake ?

Thanks a lot :)

Marie


On Tue, Oct 4, 2011 at 7:24 PM, Jeremy Selan <jeremy...@...> wrote:
Marie,

Thanks for catching this!   The currently shipping IIF configuration
is based on an older specification (I forget the exact version, but
it's around 8 months old), so my hope is the versions will explain the
differences you are seeing.  We've wanted to update to the latest RRT
for awhile now, so let us take a stab at it and we'll see if it fixes
everything.  We will also take care to note in the profile which RRT
version it is specifically, to help avoid ambiguity in the future.

The noise (discontinuity) you are seeing in the 3d RRT lut is an
artifact of how we originally generated this table, we'll make sure
it's all fixed in the updated version.

The AllocationTransform you mention describes a linear -> log
transform (a perfect mathematical log operator), where the range from
( 2^-10.4739 , 2^5.52607)  is rolled into (0.0-1.0) for sampling with
a 3d lut.  The lut was generated by taking a 3d lattice image (0-1) ,
unwarping it through the inverse of the transform, and then applying
the analytical rrt.

Thanks!

-- Jeremy


2011/10/4 Marie Fétiveau <m...@...>:
> Hello all !
> I'm having a look at IIF config.ocio and I was wondering how you build
> the aceslg2_to_Rec709.cube LUT for the rrt_odt_r709 colorspace.
> I'm asking because when I compare an OCIOColorspace node (in : aces, out :
> rrt_odt_r709) and two TuttleCTL nodes (one with the RRT 2.2.1 and another
> with the REC701 softproof ODT), there are some differences :
> - a small shift
> - + some noise in yellows with OCIOColorspace node.
> Here's a screenshot of my nuke scene and a jpg export with TuttleCTL nodes
> and an OpenColorIO Node.
> For my tests, I'm using an XRite colorchart shot with a Red One and exported
> in EXR thanks to REDCineX (ACES option enabled).
> am I mis-using the OCIOColorspace node ?
> Anyway, I will be pleased to understand how the LUT was processed and
> configured in the ocio config file.
> I don't fully understand the purpose of the AllocationTransform vars before
> the FileTransform :
> - !<ColorSpace>
>     name: rrt_odt_r709
>     family: rrt_odt_r709
>     bitdepth: 32f
>     isdata: false
>     allocation: uniform
>     allocationvars: [0, 1]
>     from_reference: !<GroupTransform>
>       children:
>         - !<AllocationTransform> {allocation: lg2, vars: [-10.4739,
> 5.52607]}
>         - !<FileTransform> {src: aceslg2_to_Rec709.cube, interpolation:
> linear}
>
>
> Thanks a lot !
> Marie
Jeremy Selan <jeremy...@...>
 

Your example looks good to me. Can you test the sub components of
your transform?

If you skip the ctl nodes, and only try the log2lin portions, does
that result in an identity transform?

Feel free to send me your lut files / .nk files offline, I'd be happy
to take a look as well.

-- Jeremy

2011/10/5 Marie Fétiveau <m...@...>:

Thanks for the quick answer and the explanation :)
I tried this afternoon to create my own aceslg2_to_Rec709 LUT but I'm still
not sure to understand how it should work.
I started to give a quick try with the Nuke Log2Lin node to see what
happened.
So I connected a CMSTestPattern + a Nuke LogToLin node + my 2  CTL node (RRT
+ ODT) + a generate LUT to create a logToLin_RRT_ODTr709.cube.
I also create a linToLog.cube (which correspond to Nuke LinToLog).
And I set the rrt_odt_r709 colorspace like this :
from_reference: !<GroupTransform>
      children:
        - !<FileTransform> {src: linToLog.cube, interpolation: linear}
        - !<FileTransform> {src: logToLin_RRT_ODTr709.cube, interpolation:
linear}
When I compare the two CTL nodes and the OCIOColorspace, that's not that bad
but there's a kind of offset + gamma shift...
 A picture is often better than a long talk away (I put the viewer in sRGB
instead linear to emphasize the shift on the screenshot).
Where's my mistake ?
Thanks a lot :)
Marie

On Tue, Oct 4, 2011 at 7:24 PM, Jeremy Selan <jeremy...@...> wrote:

Marie,

Thanks for catching this!   The currently shipping IIF configuration
is based on an older specification (I forget the exact version, but
it's around 8 months old), so my hope is the versions will explain the
differences you are seeing.  We've wanted to update to the latest RRT
for awhile now, so let us take a stab at it and we'll see if it fixes
everything.  We will also take care to note in the profile which RRT
version it is specifically, to help avoid ambiguity in the future.

The noise (discontinuity) you are seeing in the 3d RRT lut is an
artifact of how we originally generated this table, we'll make sure
it's all fixed in the updated version.

The AllocationTransform you mention describes a linear -> log
transform (a perfect mathematical log operator), where the range from
( 2^-10.4739 , 2^5.52607)  is rolled into (0.0-1.0) for sampling with
a 3d lut.  The lut was generated by taking a 3d lattice image (0-1) ,
unwarping it through the inverse of the transform, and then applying
the analytical rrt.

Thanks!

-- Jeremy


2011/10/4 Marie Fétiveau <m...@...>:
Hello all !
I'm having a look at IIF config.ocio and I was wondering how you build
the aceslg2_to_Rec709.cube LUT for the rrt_odt_r709 colorspace.
I'm asking because when I compare an OCIOColorspace node (in : aces, out
:
rrt_odt_r709) and two TuttleCTL nodes (one with the RRT 2.2.1 and
another
with the REC701 softproof ODT), there are some differences :
- a small shift
- + some noise in yellows with OCIOColorspace node.
Here's a screenshot of my nuke scene and a jpg export
with TuttleCTL nodes
and an OpenColorIO Node.
For my tests, I'm using an XRite colorchart shot with a Red One and
exported
in EXR thanks to REDCineX (ACES option enabled).
am I mis-using the OCIOColorspace node ?
Anyway, I will be pleased to understand how the LUT was processed and
configured in the ocio config file.
I don't fully understand the purpose of the AllocationTransform vars
before
the FileTransform :
- !<ColorSpace>
    name: rrt_odt_r709
    family: rrt_odt_r709
    bitdepth: 32f
    isdata: false
    allocation: uniform
    allocationvars: [0, 1]
    from_reference: !<GroupTransform>
      children:
        - !<AllocationTransform> {allocation: lg2, vars: [-10.4739,
5.52607]}
        - !<FileTransform> {src: aceslg2_to_Rec709.cube, interpolation:
linear}


Thanks a lot !
Marie
Jeremy Selan <jeremy...@...>
 

I just rolled out a new iif profile, with updated 3dluts.

You'll note that inside the iif profile (in the lutimages subdir) are
the lattice images used to bake the rrt / odt into 3d luts. So you
should be able to use these image to bake any rrts you may be using
into a lut suitable for inclusion in the profile.

Please let me know if, using this updated approach, you are not able
to recreate an identical result.

-- Jeremy

On Mon, Oct 10, 2011 at 2:58 PM, Jeremy Selan <jeremy...@...> wrote:
Your example looks good to me.  Can you test the sub components of
your transform?

If you skip the ctl nodes, and only try the log2lin portions, does
that result in an identity transform?

Feel free to send me your lut files / .nk files offline, I'd be happy
to take a look as well.

-- Jeremy

2011/10/5 Marie Fétiveau <m...@...>:
Thanks for the quick answer and the explanation :)
I tried this afternoon to create my own aceslg2_to_Rec709 LUT but I'm still
not sure to understand how it should work.
I started to give a quick try with the Nuke Log2Lin node to see what
happened.
So I connected a CMSTestPattern + a Nuke LogToLin node + my 2  CTL node (RRT
+ ODT) + a generate LUT to create a logToLin_RRT_ODTr709.cube.
I also create a linToLog.cube (which correspond to Nuke LinToLog).
And I set the rrt_odt_r709 colorspace like this :
from_reference: !<GroupTransform>
      children:
        - !<FileTransform> {src: linToLog.cube, interpolation: linear}
        - !<FileTransform> {src: logToLin_RRT_ODTr709.cube, interpolation:
linear}
When I compare the two CTL nodes and the OCIOColorspace, that's not that bad
but there's a kind of offset + gamma shift...
 A picture is often better than a long talk away (I put the viewer in sRGB
instead linear to emphasize the shift on the screenshot).
Where's my mistake ?
Thanks a lot :)
Marie

On Tue, Oct 4, 2011 at 7:24 PM, Jeremy Selan <jeremy...@...> wrote:

Marie,

Thanks for catching this!   The currently shipping IIF configuration
is based on an older specification (I forget the exact version, but
it's around 8 months old), so my hope is the versions will explain the
differences you are seeing.  We've wanted to update to the latest RRT
for awhile now, so let us take a stab at it and we'll see if it fixes
everything.  We will also take care to note in the profile which RRT
version it is specifically, to help avoid ambiguity in the future.

The noise (discontinuity) you are seeing in the 3d RRT lut is an
artifact of how we originally generated this table, we'll make sure
it's all fixed in the updated version.

The AllocationTransform you mention describes a linear -> log
transform (a perfect mathematical log operator), where the range from
( 2^-10.4739 , 2^5.52607)  is rolled into (0.0-1.0) for sampling with
a 3d lut.  The lut was generated by taking a 3d lattice image (0-1) ,
unwarping it through the inverse of the transform, and then applying
the analytical rrt.

Thanks!

-- Jeremy


2011/10/4 Marie Fétiveau <m...@...>:
Hello all !
I'm having a look at IIF config.ocio and I was wondering how you build
the aceslg2_to_Rec709.cube LUT for the rrt_odt_r709 colorspace.
I'm asking because when I compare an OCIOColorspace node (in : aces, out
:
rrt_odt_r709) and two TuttleCTL nodes (one with the RRT 2.2.1 and
another
with the REC701 softproof ODT), there are some differences :
- a small shift
- + some noise in yellows with OCIOColorspace node.
Here's a screenshot of my nuke scene and a jpg export
with TuttleCTL nodes
and an OpenColorIO Node.
For my tests, I'm using an XRite colorchart shot with a Red One and
exported
in EXR thanks to REDCineX (ACES option enabled).
am I mis-using the OCIOColorspace node ?
Anyway, I will be pleased to understand how the LUT was processed and
configured in the ocio config file.
I don't fully understand the purpose of the AllocationTransform vars
before
the FileTransform :
- !<ColorSpace>
    name: rrt_odt_r709
    family: rrt_odt_r709
    bitdepth: 32f
    isdata: false
    allocation: uniform
    allocationvars: [0, 1]
    from_reference: !<GroupTransform>
      children:
        - !<AllocationTransform> {allocation: lg2, vars: [-10.4739,
5.52607]}
        - !<FileTransform> {src: aceslg2_to_Rec709.cube, interpolation:
linear}


Thanks a lot !
Marie
Marie Fétiveau <m...@...>
 

Great :)
That's much much better.
Thanks a lot !
I still have some small differencies comparing OCIOColorspaceNode and my two TuttleCTL nodes.
But I need to compare with CTLRender just to be sure it's not a pb with TuttleCTL.

I have another pb. When I export a 3DLUT thanks to ociobakeLut and compare the resulting LUT (using a VectorField node) with the output of a OCIOColorspace node, there a noticable shift. 
Here's my command : ociobakelut --inputspace aces --outputspace rrt_sRGB --format flame aces_to_ODT_sRGB.3dl

Assuming it's a LUT approximation problem, I tried to export 32, 64 and 128 segments LUTs. And indeed, the more I add segments, the less the shift is important.
But even at 128 segments (which is huge), there's still a tiny shift.

May be the RRT is so complex that a linear sampled LUT couldn't be enough ?

Anyway, on the way out I noticed a bug in ociobakeLut using the --cubesize option and exporting 3dl : the header is always computed for 17 segments.
For exemple with cubesize=32 :
header = 0 64 128 192 256 320 384 448 512 575 639 703 767 831 895 959 1023
should be = 0   33   66   99   132   165   198   231   264   297   330   363   396   429   462   495   528   561   594   627   660   693   726   759   792   825   858   891   924   957   990   1023

Thanks again.

Marie


On Wed, Oct 26, 2011 at 12:11 AM, Jeremy Selan <jeremy...@...> wrote:
I just rolled out a new iif profile, with updated 3dluts.

You'll note that inside the iif profile (in the lutimages subdir) are
the lattice images used to bake the rrt / odt into 3d luts. So you
should be able to use these image to bake any rrts you may be using
into a lut suitable for inclusion in the profile.

Please let me know if, using this updated approach, you are not able
to recreate an identical result.

-- Jeremy

On Mon, Oct 10, 2011 at 2:58 PM, Jeremy Selan <jeremy...@...> wrote:
> Your example looks good to me.  Can you test the sub components of
> your transform?
>
> If you skip the ctl nodes, and only try the log2lin portions, does
> that result in an identity transform?
>
> Feel free to send me your lut files / .nk files offline, I'd be happy
> to take a look as well.
>
> -- Jeremy
>
> 2011/10/5 Marie Fétiveau <m...@...>:
>> Thanks for the quick answer and the explanation :)
>> I tried this afternoon to create my own aceslg2_to_Rec709 LUT but I'm still
>> not sure to understand how it should work.
>> I started to give a quick try with the Nuke Log2Lin node to see what
>> happened.
>> So I connected a CMSTestPattern + a Nuke LogToLin node + my 2  CTL node (RRT
>> + ODT) + a generate LUT to create a logToLin_RRT_ODTr709.cube.
>> I also create a linToLog.cube (which correspond to Nuke LinToLog).
>> And I set the rrt_odt_r709 colorspace like this :
>> from_reference: !<GroupTransform>
>>       children:
>>         - !<FileTransform> {src: linToLog.cube, interpolation: linear}
>>         - !<FileTransform> {src: logToLin_RRT_ODTr709.cube, interpolation:
>> linear}
>> When I compare the two CTL nodes and the OCIOColorspace, that's not that bad
>> but there's a kind of offset + gamma shift...
>>  A picture is often better than a long talk away (I put the viewer in sRGB
>> instead linear to emphasize the shift on the screenshot).
>> Where's my mistake ?
>> Thanks a lot :)
>> Marie
>>
>> On Tue, Oct 4, 2011 at 7:24 PM, Jeremy Selan <jeremy...@...> wrote:
>>>
>>> Marie,
>>>
>>> Thanks for catching this!   The currently shipping IIF configuration
>>> is based on an older specification (I forget the exact version, but
>>> it's around 8 months old), so my hope is the versions will explain the
>>> differences you are seeing.  We've wanted to update to the latest RRT
>>> for awhile now, so let us take a stab at it and we'll see if it fixes
>>> everything.  We will also take care to note in the profile which RRT
>>> version it is specifically, to help avoid ambiguity in the future.
>>>
>>> The noise (discontinuity) you are seeing in the 3d RRT lut is an
>>> artifact of how we originally generated this table, we'll make sure
>>> it's all fixed in the updated version.
>>>
>>> The AllocationTransform you mention describes a linear -> log
>>> transform (a perfect mathematical log operator), where the range from
>>> ( 2^-10.4739 , 2^5.52607)  is rolled into (0.0-1.0) for sampling with
>>> a 3d lut.  The lut was generated by taking a 3d lattice image (0-1) ,
>>> unwarping it through the inverse of the transform, and then applying
>>> the analytical rrt.
>>>
>>> Thanks!
>>>
>>> -- Jeremy
>>>
>>>
>>> 2011/10/4 Marie Fétiveau <m...@...>:
>>> > Hello all !
>>> > I'm having a look at IIF config.ocio and I was wondering how you build
>>> > the aceslg2_to_Rec709.cube LUT for the rrt_odt_r709 colorspace.
>>> > I'm asking because when I compare an OCIOColorspace node (in : aces, out
>>> > :
>>> > rrt_odt_r709) and two TuttleCTL nodes (one with the RRT 2.2.1 and
>>> > another
>>> > with the REC701 softproof ODT), there are some differences :
>>> > - a small shift
>>> > - + some noise in yellows with OCIOColorspace node.
>>> > Here's a screenshot of my nuke scene and a jpg export
>>> > with TuttleCTL nodes
>>> > and an OpenColorIO Node.
>>> > For my tests, I'm using an XRite colorchart shot with a Red One and
>>> > exported
>>> > in EXR thanks to REDCineX (ACES option enabled).
>>> > am I mis-using the OCIOColorspace node ?
>>> > Anyway, I will be pleased to understand how the LUT was processed and
>>> > configured in the ocio config file.
>>> > I don't fully understand the purpose of the AllocationTransform vars
>>> > before
>>> > the FileTransform :
>>> > - !<ColorSpace>
>>> >     name: rrt_odt_r709
>>> >     family: rrt_odt_r709
>>> >     bitdepth: 32f
>>> >     isdata: false
>>> >     allocation: uniform
>>> >     allocationvars: [0, 1]
>>> >     from_reference: !<GroupTransform>
>>> >       children:
>>> >         - !<AllocationTransform> {allocation: lg2, vars: [-10.4739,
>>> > 5.52607]}
>>> >         - !<FileTransform> {src: aceslg2_to_Rec709.cube, interpolation:
>>> > linear}
>>> >
>>> >
>>> > Thanks a lot !
>>> > Marie
>>
>>
>
dbr/Ben <dbr....@...>
 

On 28/10/2011, at 1:15, Marie Fétiveau <m...@...> wrote:

compare the resulting LUT (using a VectorField node) with the output of a OCIOColorspace node, there a noticable shift.
The Vectorfield node is a bit buggy - even if you write out a no-op lut (CMS test pattern->GenerateLUT) you get a luminance shift with several formats:

http://opencolorio.org/FAQ.html#what-are-the-differences-between-nuke-s-vectorfield-and-ociofiletransform

Anyway, on the way out I noticed a bug in ociobakeLut using the --cubesize option and exporting 3dl : the header is always computed for 17 segments.
Think the "ociobakut --format flame" and lustre 3dl bakers have fixed cube sizes, as I guess the applications only accept these sized LUT's?

If that's true, would be easy to add a more generic --format 3dl that respects the cube size (and have the flame/lustre formats error usefully of you try and change the cube size)
Marie Fétiveau <m...@...>
 

Thanks for the reply !

The Vectorfield node is a bit buggy - even if you write out a no-op lut (CMS test pattern->GenerateLUT) you get a luminance shift with several formats:

http://opencolorio.org/FAQ.html#what-are-the-differences-between-nuke-s-vectorfield-and-ociofiletransform
Ok, I didn't know that. I'll have a look, it's interesting !
I see in the table that 3dl are ok... But I'm using Nuke 6.2v4...

And I also have a shift with RV. Has RV a pb with 3dls too ?

Can you advice me a soft where a can surely check my LUT ?

Think the "ociobakut --format flame" and lustre 3dl bakers have fixed cube sizes, as I guess the applications only accept these sized LUT's?
Ok, I didn't figure out what were the particularities of "lustre" and "flame" 3dl formats. That makes sense. I needed a 3dl so I used by default the flame format.

But it is still strange to me : when using the --cubesize 32, the generated LUT is really on 32 segments. Only the header is on 17 segments. If I replace the 17 header by a 32 header, the LUT works on Nuke and RV (with a 17 header, I get weird colors).
Does that mean that Flame, always need a 17 header but can neverthemless parse other size LUTs ?

thanks !

Marie
Jeremy Selan <jeremy...@...>
 

This feels like a bug (not having the shaper lut obey the specified size).

If I remember correctly (the last time I tested at least) flame only
supportex size 17 luts, so even if we make the fix it wont effect
that. The real test would be to validate the lustre can load a 3dl
with a sized 32 header.

-- Jeremy


2011/10/28 Marie Fétiveau <m...@...>:

Ok, I didn't figure out what were the particularities of "lustre" and
"flame" 3dl formats. That makes sense. I needed a 3dl so I used by default
the flame format.
But it is still strange to me : when using the --cubesize 32, the generated
LUT is really on 32 segments. Only the header is on 17 segments. If I
replace the 17 header by a 32 header, the LUT works on Nuke and RV (with a
17 header, I get weird colors).
Does that mean that Flame, always need a 17 header but can neverthemless
parse other size LUTs ?
thanks !
Marie
Jeremy Selan <jeremy...@...>
 

A fix for this was just committed to master:
https://github.com/imageworks/OpenColorIO/pull/177

Please let me know if you have any other compatibility issues with 3dl export.

-- Jeremy

On Mon, Oct 31, 2011 at 11:14 AM, Jeremy Selan <jeremy...@...> wrote:
This feels like a bug (not having the shaper lut obey the specified size).

If I remember correctly (the last time I tested at least) flame only
supportex size 17 luts, so even if we make the fix it wont effect
that.  The real test would be to validate the lustre can load a 3dl
with a sized 32 header.

-- Jeremy


2011/10/28 Marie Fétiveau <m...@...>:
Ok, I didn't figure out what were the particularities of "lustre" and
"flame" 3dl formats. That makes sense. I needed a 3dl so I used by default
the flame format.
But it is still strange to me : when using the --cubesize 32, the generated
LUT is really on 32 segments. Only the header is on 17 segments. If I
replace the 17 header by a 32 header, the LUT works on Nuke and RV (with a
17 header, I get weird colors).
Does that mean that Flame, always need a 17 header but can neverthemless
parse other size LUTs ?
thanks !
Marie
Joseph Slomka <jsl...@...>
 

Marie,
 
I am VERY late to the party but a linear spaced 32 or 64 segment lut from aces to a ssrt/odt will be insufficent.
You can create a ACES ADX-> rrt/odt  32 or 64 lut and get acceptable results.  
 
 
-Joseph


From: ocio...@... [mailto:ocio...@...] On Behalf Of Marie Fétiveau
Sent: Thursday, October 27, 2011 7:45 AM
To: ocio...@...
Subject: Re: [ocio-dev] Question about IIF config

Great :)
That's much much better.
Thanks a lot !
I still have some small differencies comparing OCIOColorspaceNode and my two TuttleCTL nodes.
But I need to compare with CTLRender just to be sure it's not a pb with TuttleCTL.

I have another pb. When I export a 3DLUT thanks to ociobakeLut and compare the resulting LUT (using a VectorField node) with the output of a OCIOColorspace node, there a noticable shift. 
Here's my command : ociobakelut --inputspace aces --outputspace rrt_sRGB --format flame aces_to_ODT_sRGB.3dl

Assuming it's a LUT approximation problem, I tried to export 32, 64 and 128 segments LUTs. And indeed, the more I add segments, the less the shift is important.
But even at 128 segments (which is huge), there's still a tiny shift.

May be the RRT is so complex that a linear sampled LUT couldn't be enough ?

Anyway, on the way out I noticed a bug in ociobakeLut using the --cubesize option and exporting 3dl : the header is always computed for 17 segments.
For exemple with cubesize=32 :
header = 0 64 128 192 256 320 384 448 512 575 639 703 767 831 895 959 1023
should be = 0   33   66   99   132   165   198   231   264   297   330   363   396   429   462   495   528   561   594   627   660   693   726   759   792   825   858   891   924   957   990   1023

Thanks again.

Marie

On Wed, Oct 26, 2011 at 12:11 AM, Jeremy Selan <jeremy...@...> wrote:
I just rolled out a new iif profile, with updated 3dluts.

You'll note that inside the iif profile (in the lutimages subdir) are
the lattice images used to bake the rrt / odt into 3d luts. So you
should be able to use these image to bake any rrts you may be using
into a lut suitable for inclusion in the profile.

Please let me know if, using this updated approach, you are not able
to recreate an identical result.

-- Jeremy

On Mon, Oct 10, 2011 at 2:58 PM, Jeremy Selan <jeremy...@...> wrote:
> Your example looks good to me.  Can you test the sub components of
> your transform?
>
> If you skip the ctl nodes, and only try the log2lin portions, does
> that result in an identity transform?
>
> Feel free to send me your lut files / .nk files offline, I'd be happy
> to take a look as well.
>
> -- Jeremy
>
> 2011/10/5 Marie Fétiveau <m...@...>:
>> Thanks for the quick answer and the explanation :)
>> I tried this afternoon to create my own aceslg2_to_Rec709 LUT but I'm still
>> not sure to understand how it should work.
>> I started to give a quick try with the Nuke Log2Lin node to see what
>> happened.
>> So I connected a CMSTestPattern + a Nuke LogToLin node + my 2  CTL node (RRT
>> + ODT) + a generate LUT to create a logToLin_RRT_ODTr709.cube.
>> I also create a linToLog.cube (which correspond to Nuke LinToLog).
>> And I set the rrt_odt_r709 colorspace like this :
>> from_reference: !<GroupTransform>
>>       children:
>>         - !<FileTransform> {src: linToLog.cube, interpolation: linear}
>>         - !<FileTransform> {src: logToLin_RRT_ODTr709.cube, interpolation:
>> linear}
>> When I compare the two CTL nodes and the OCIOColorspace, that's not that bad
>> but there's a kind of offset + gamma shift...
>>  A picture is often better than a long talk away (I put the viewer in sRGB
>> instead linear to emphasize the shift on the screenshot).
>> Where's my mistake ?
>> Thanks a lot :)
>> Marie
>>
>> On Tue, Oct 4, 2011 at 7:24 PM, Jeremy Selan <jeremy...@...> wrote:
>>>
>>> Marie,
>>>
>>> Thanks for catching this!   The currently shipping IIF configuration
>>> is based on an older specification (I forget the exact version, but
>>> it's around 8 months old), so my hope is the versions will explain the
>>> differences you are seeing.  We've wanted to update to the latest RRT
>>> for awhile now, so let us take a stab at it and we'll see if it fixes
>>> everything.  We will also take care to note in the profile which RRT
>>> version it is specifically, to help avoid ambiguity in the future.
>>>
>>> The noise (discontinuity) you are seeing in the 3d RRT lut is an
>>> artifact of how we originally generated this table, we'll make sure
>>> it's all fixed in the updated version.
>>>
>>> The AllocationTransform you mention describes a linear -> log
>>> transform (a perfect mathematical log operator), where the range from
>>> ( 2^-10.4739 , 2^5.52607)  is rolled into (0.0-1.0) for sampling with
>>> a 3d lut.  The lut was generated by taking a 3d lattice image (0-1) ,
>>> unwarping it through the inverse of the transform, and then applying
>>> the analytical rrt.
>>>
>>> Thanks!
>>>
>>> -- Jeremy
>>>
>>>
>>> 2011/10/4 Marie Fétiveau <m...@...>:
>>> > Hello all !
>>> > I'm having a look at IIF config.ocio and I was wondering how you build
>>> > the aceslg2_to_Rec709.cube LUT for the rrt_odt_r709 colorspace.
>>> > I'm asking because when I compare an OCIOColorspace node (in : aces, out
>>> > :
>>> > rrt_odt_r709) and two TuttleCTL nodes (one with the RRT 2.2.1 and
>>> > another
>>> > with the REC701 softproof ODT), there are some differences :
>>> > - a small shift
>>> > - + some noise in yellows with OCIOColorspace node.
>>> > Here's a screenshot of my nuke scene and a jpg export
>>> > with TuttleCTL nodes
>>> > and an OpenColorIO Node.
>>> > For my tests, I'm using an XRite colorchart shot with a Red One and
>>> > exported
>>> > in EXR thanks to REDCineX (ACES option enabled).
>>> > am I mis-using the OCIOColorspace node ?
>>> > Anyway, I will be pleased to understand how the LUT was processed and
>>> > configured in the ocio config file.
>>> > I don't fully understand the purpose of the AllocationTransform vars
>>> > before
>>> > the FileTransform :
>>> > - !<ColorSpace>
>>> >     name: rrt_odt_r709
>>> >     family: rrt_odt_r709
>>> >     bitdepth: 32f
>>> >     isdata: false
>>> >     allocation: uniform
>>> >     allocationvars: [0, 1]
>>> >     from_reference: !<GroupTransform>
>>> >       children:
>>> >         - !<AllocationTransform> {allocation: lg2, vars: [-10.4739,
>>> > 5.52607]}
>>> >         - !<FileTransform> {src: aceslg2_to_Rec709.cube, interpolation:
>>> > linear}
>>> >
>>> >
>>> > Thanks a lot !
>>> > Marie
>>
>>
>
1 - 10 of 10