Extra Buffer nodes
- Thread starter dpont
- Start date
Gregg "T.Rex"
Carnotaur.Rex
Hey Greg! how are you doing?
Yes, that's actually something I was thinking as well. All those things are implementable in compositing packages if the proper buffers are exported - even spot replacement effects can be implemented with a wp pass, but the advantage of solving some of those things at pixel level at render time should be noted too.
For example, a very shallow Depth of Field filter applied at pixel level in render time offers better results than applying it at image level within a compositing package. Reason for this is we are able to filter some few 3D DOF passes and get data from what's behind the foreground geometry without rendering 2 layers (FG/BG) by separate, so results are not only better but also we can use less AA passes than solving it with 3D DOF solely. You'll probably remember X-Dof worked in that way. In similar way, it's possible yet with 2 instances of DP NPF. Being able to do it in node editor allows to configure some things in a flexible way without any programming skills. There's also other scenarios where we might want to render at a specific reference white and perform post white balancing according to another illuminant. Settings things up at pixel filter level may be more flexible for changing easily the reference white, instead of changing HDR lightprobe and each lights separately. Some other effects would be better applied after surface evaluation like a kind of global shader node editor, I guess.
Good thing is that having these things working at both pixel and image filter level allows the user to have some boosted degree of post-processing power within LightWave. So for quick concept designs, look development, storyboardings, animatics, art previz or just for selling quick preliminary ideas to a client, solving all these render things within a single package is really very handy.
Cheers!
Gerardo
p.s. Just noticed in case you were referring about the specific case of the method shared here, only the second filter is implementable in compositing package, easier with a macro or gizmo.
Yes, node compositing at pixel level before rendering has a ton of advantages, comparing to working with just the resulted raster images.
Katana and Clarisse are very good at that, Nuke can use Vray to render 3d scenes directly in comp, but LW had these pixel filtering node tools way before everyone else.
I was referring to the fireflies/overblown example you posted.
No matter the app used, this involves kind of a tone mapping process, one way or another, before and during the rendering process.
I wonder if there is a way to just render raw burned images, maybe along with some required passes and get results back in post working at raster level.
Cheers!
Greg
jeric_synergy
Axes grinder- Dongle #99
:stumped: You lost me there: it reads (to me) like you're contradicting yourself-- what am I missing???
:thumbsup:
gerardstrada
New member
hmmm... that's an interesting question. What I would try is to find first higher luminance values, then isolate them with a high frequency filter to find the spots only and then use that as mask to tone map selectively the problematic areas. But the problem to try something like that is the secondary reflection bounces. Because those will still illuminate little bright spots in diffusion component that a filter that only addresses super bright spots in post can not solve. Guess after finding the little bright spots, a second pass would be necessary to expand the kernel of the problematic areas around the bright spots (I'm guessing some kind of distance gradient based on normals as you imply as required pass might help). But still, the color contribution of those areas still would be there. Guess it should be a way, but have the idea it would be more tricky than the proposed method
Cheers!
Gerardo
p.s. Jeric, sorry don't understand what part is confusing you. If you could elaborate maybe I might help.
Gregg "T.Rex"
Carnotaur.Rex
hmmm... that's an interesting question. What I would try is to find first higher luminance values, then isolate them with a high frequency filter to find the spots only and then use that as mask to tone map selectively the problematic areas. But the problem to try something like that is the secondary reflection bounces. Because those will still illuminate little bright spots in diffusion component that a filter that only addresses super bright spots in post can not solve. Guess after finding the little bright spots, a second pass would be necessary to expand the kernel of the problematic areas around the bright spots (I'm guessing some kind of distance gradient based on normals as you imply as required pass might help). But still, the color contribution of those areas still would be there. Guess it should be a way, but have the idea it would be more tricky than the proposed method
Cheers!
Gerardo
Heh...
I was hoping for a more "appropriate" solution, sort of what you did with pixel filter, not hacking/masking our way through the blown areas, but i guess you're right; this problem has to be dealt before/during rendering and not in post...
Thanks, always appreciate your input!
Cheers,
Greg
gerardstrada
New member
If you could elaborate maybe I might help with the part is confusing you
Gerardo
gerardstrada
New member
Yes, the highlight compression part is better addressed at render time. If you are using Vray within Nuke, maybe you want to try with Exponential Color Mapping and then apply the inverse formula I shared for DP NIF through a gizmo.
Hope it works!
Cheers,
Gerardo
Pavlov
- pavlov who ? -
Hi all,
since discussion came up to life again, i jump in to ask a question on something i'm struggling on for a while.
I use to export BW masks of materials and objects as separate files to use them in PS as masks. Now i use to do this manually, using matte object option for object, and setting up surfaces manually as black or 100% white to get material masks.
Is there a setup which allows me to save out BW masks for given object / surfaces automatically ?
Best
Paolo
since discussion came up to life again, i jump in to ask a question on something i'm struggling on for a while.
I use to export BW masks of materials and objects as separate files to use them in PS as masks. Now i use to do this manually, using matte object option for object, and setting up surfaces manually as black or 100% white to get material masks.
Is there a setup which allows me to save out BW masks for given object / surfaces automatically ?
Best
Paolo
gerardstrada
New member
Yep, as gar26lw says, DP Mask Objects/Surfaces nodes can be used in DP NPF.
You can plug them in Scalar Global Buffer or you can also plug 3 different masks in RGB Color Global Buffers (so a single RGB image would contain 3 different masks you can extract later in Ps). For Surfaces you could use also ExtraBuffers node maybe. Be sure you have last version of DP Filters.
Gerardo
You can plug them in Scalar Global Buffer or you can also plug 3 different masks in RGB Color Global Buffers (so a single RGB image would contain 3 different masks you can extract later in Ps). For Surfaces you could use also ExtraBuffers node maybe. Be sure you have last version of DP Filters.
Gerardo
gerardstrada
New member
Just in case some people don't know, Denis Pontonnier has updated DP Filter Node Editors for LW 2018!!! There are some new things, but before addressing those topics, I wanted to try the setup shared here to avoid bright spot artifacts without losing dynamic range. My previous images can not be seen but they are still there (server has changed its domain from .org to .cc), so let's try how this technique behaves in LW 2018.
High contrast ratios in illumination tend to produce artifacts of super bright pixels due to MC sampling of glossy reflective rays in some materials in LW2018. Artifacts from direct specular can be sometimes avoided by increasing Reflection Recursion Limit, but this also may increase render times a lot and it won't solve indirect specular artifacts. In LW 2018 these artifacts may look like this:
Those artifacts could be solved by some of these methods or probably with MultiImportanceSampling on indirect rays.
At user level however the common route is to clip values at render times, but this produces inconveniences like lost of HDR data and unrealistic highlights when post-processing the image:
and the worst thing is that sometimes we can still get some clipped isolated spots there in v2018. We can try Median filters with some clever masking techniques in post-processing to help to remove hot pixels
but sometimes we can lose also details definition or could not get rid of all those super-bright pixels.
The proposed technique works in the same way in v2018 and could be used in conjunction with median filters, or even alone when treating still images. The method used a simple TMO (Reinhard) in DP PFNE:
the proposed node setup looked like this:
applied to the luminance component at render time to keep those pixels at a maximum of 1.0 or below:
Tip: We can get real time preview of this result in VPR.
Then the inverse formula is re-applied at post-processing in DP IFNE:
to return back the proper dynamic range in the highlights.
Tip: We can preview what we do at post-processing by enabling IFNE Viewer! Just double click in ImageFilter node to access this option.
Gerardo
High contrast ratios in illumination tend to produce artifacts of super bright pixels due to MC sampling of glossy reflective rays in some materials in LW2018. Artifacts from direct specular can be sometimes avoided by increasing Reflection Recursion Limit, but this also may increase render times a lot and it won't solve indirect specular artifacts. In LW 2018 these artifacts may look like this:
Those artifacts could be solved by some of these methods or probably with MultiImportanceSampling on indirect rays.
At user level however the common route is to clip values at render times, but this produces inconveniences like lost of HDR data and unrealistic highlights when post-processing the image:
and the worst thing is that sometimes we can still get some clipped isolated spots there in v2018. We can try Median filters with some clever masking techniques in post-processing to help to remove hot pixels
but sometimes we can lose also details definition or could not get rid of all those super-bright pixels.
The proposed technique works in the same way in v2018 and could be used in conjunction with median filters, or even alone when treating still images. The method used a simple TMO (Reinhard) in DP PFNE:
the proposed node setup looked like this:
applied to the luminance component at render time to keep those pixels at a maximum of 1.0 or below:
Tip: We can get real time preview of this result in VPR.
Then the inverse formula is re-applied at post-processing in DP IFNE:
to return back the proper dynamic range in the highlights.
Tip: We can preview what we do at post-processing by enabling IFNE Viewer! Just double click in ImageFilter node to access this option.
Gerardo
Last edited:
gerardstrada
New member
Result looks something like this in LW2018:
Nothing bad, eh? But sometimes we need to use even higher contrast ratios. In this case the contrast ratio is 3x times higher than the previous case:
The same v2015 setup gives us this result in LW2018.
Which is fine for stills, but for animations some spots could pop up.
Again, this solution could be mixed with Median filters in post to get rid of those fireflies, but what if we would like to solve it as much as possible at render time? This new setup applies a flatten curve besides the soft-clipping in order to remove those rebel fireflies:
Luminance is optimized for Rec709/sRGB space in this case, so if you are using other colorspace, you'll need to change those values. We can try with any logarithmic curve used for post-processing (LogC, ACEScc, REDLog, PanaLog, etc..) and also Cineon curve from the new native ConvertColorSpace node works as well:
In DP IFNE we expand back to the linear curve:
Result looks like this:
Note: with a logarithmic curve, AA edges will look different than with the soft-clipping method.
As we see, the only firefly there is the one we have modeled The technique is not fully tested in v2018 but you might find it useful in some cases ...attaching the compound nodes of the previous and the new setup.
Merry Christmas!
Gerardo
p.s. no firefly has been harmed in the realization of these tests
Nothing bad, eh? But sometimes we need to use even higher contrast ratios. In this case the contrast ratio is 3x times higher than the previous case:
The same v2015 setup gives us this result in LW2018.
Which is fine for stills, but for animations some spots could pop up.
Again, this solution could be mixed with Median filters in post to get rid of those fireflies, but what if we would like to solve it as much as possible at render time? This new setup applies a flatten curve besides the soft-clipping in order to remove those rebel fireflies:
Luminance is optimized for Rec709/sRGB space in this case, so if you are using other colorspace, you'll need to change those values. We can try with any logarithmic curve used for post-processing (LogC, ACEScc, REDLog, PanaLog, etc..) and also Cineon curve from the new native ConvertColorSpace node works as well:
In DP IFNE we expand back to the linear curve:
Result looks like this:
Note: with a logarithmic curve, AA edges will look different than with the soft-clipping method.
As we see, the only firefly there is the one we have modeled
The technique is not fully tested in v2018 but you might find it useful in some cases ...attaching the compound nodes of the previous and the new setup.Merry Christmas!
Gerardo
p.s. no firefly has been harmed in the realization of these tests
Last edited:
gerardstrada
New member
Ok. Denis has come up with something even better!!! ...and in record time! The new DP Median filter.
Setup can not be simpler since all is solved as post-processing just in Image Filter Node Editor. The filter works so well that you can bring to LW, renders from other 3D packages
The filter should be applied in perceptual space, or even better, in a very flat space. Something like Cineon works nicely. Setup looks like this:
So for this render:
We get this:
Or for this render:
We get this:
Think these are impressive results for so easy setup!
In both cases I'm using these settings:
Think this is gonna be one of the most used filters from now on in LW2018 if there's even cases where not all fireflies can be removed, we can still mix it with the methods shared before.
Merry Christmas!
Gerardo
Setup can not be simpler since all is solved as post-processing just in Image Filter Node Editor. The filter works so well that you can bring to LW, renders from other 3D packages
The filter should be applied in perceptual space, or even better, in a very flat space. Something like Cineon works nicely. Setup looks like this:
So for this render:
We get this:
Or for this render:
We get this:
Think these are impressive results for so easy setup!
In both cases I'm using these settings:
Think this is gonna be one of the most used filters from now on in LW2018
if there's even cases where not all fireflies can be removed, we can still mix it with the methods shared before.Merry Christmas!
Gerardo
Quantenschaum
Nonplanar Triangle
i'm stupid, so i have to ask what the settings for the 1st and the 2nd 'convert cs' nodes are... (i usually render sRGB) :-D
my current testing showed the best result when both are set to 'Linear'.
oh... did i mention how utterly brilliant this all is? :-O it's like a thorn removed from my side! THANK YOU SO MUCH !!!
my current testing showed the best result when both are set to 'Linear'.
oh... did i mention how utterly brilliant this all is? :-O it's like a thorn removed from my side! THANK YOU SO MUCH !!!
Last edited:
gerardstrada
New member
Sorry for not being clear. Using Cineon in the previous case:
tested here with a modified version of LogC and it's working nicely too so flat logarithmic curves work well
Gerardo
tested here with a modified version of LogC and it's working nicely too so flat logarithmic curves work well
Gerardo
Quantenschaum
Nonplanar Triangle
thank you very much!
Fotofashion
Photographer
Ok. Denis has come up with something even better!!! ...and in record time! The new DP Median filter.
Setup can not be simpler since all is solved as post-processing just in Image Filter Node Editor. The filter works so well that you can bring to LW, renders from other 3D packages
The filter should be applied in perceptual space, or even better, in a very flat space. Something like Cineon works nicely. Setup looks like this:
So for this render:
We get this:
Or for this render:
We get this:
Think these are impressive results for so easy setup!
In both cases I'm using these settings:
Think this is gonna be one of the most used filters from now on in LW2018
Merry Christmas!
Gerardo
This looks really nice.
Not sure how to make this work, I downloaded Dp filter for 2018, but cant find the node dp median :S is there a different plugin to download?