Tangent normal maps

[Sustaxata]

Hello everyone,

I am trying to apply some normal maps in tangent space to some models. I used both Zbrush and xNormal for exporting the normals, none of them work.

As far as I understand it, there are two types of normal maps, tangent and object space. Since the geometry will be deformed, I am using NM in tangent space.

I wanted to use the default mantra surface model first, to see how it works, but it doesn't work in the first place. It basically takes in a normal map and then uses the convertNormal subnetwork (posted below). The some of the channels still seem flipped. I tried flipping other channels using the multiply node, but it simply does not work.

I have tried looking it up, and there are some pretty ancient threats. One of them also provides an example file, but it seems houdini changed a bit since then.

http://www.sidefx.com/exchange/info.php?fileid=561&versionid=561

The VEX function

vop_computeTangents($tanU, $tanV, normalize(N), $uv, 2);

does not expect the same input anymore. I haven't really worked with VEX code yet, so I am a bit stranded here.

 

Isn't there a fast way to set tanget space normal maps up? I can't believe such a simple thing is so hard to set up.

Do I need to flip certain channels when I export the NM?

Am I overlooking something, or am I just really stupid?

 

Any kind of help or information is welcome

 

 

Thanks a lot

 

Sustaxa

[edward]

Haven't tried this before but maybe it helps:

http://www.sidefx.com/exchange/info.php?fileid=561&versionid=561

[eetu]

Ohh, I was just setting this up at work yesterday. The exchange asset had some issues; you can use the new compute tangents vop in place of the inline node, plus some vector type hints were set to 3d vector instead of 3d normal, messing things up. I'll see if I can share it on Monday, although I'm not 100% sure how standard the normals maps I was working with were.

[eetu]

And, by the way, the SideFX documentation on this is plain wrong: "Fit range from -1:1 to 0:1" should obviously be the other way round..

[eetu]

Oo-kay, this here looks like it's working, I simplified the vopnet a bit as well.

The head texture is from Nvidia. I hope the axes etc are the proper way, I don't have Zbrush here. If not, just do some more or less flipping

using_normal_maps_ee.hip

[Sustaxata]

@edward: yes thats the link, which uses the function that changed in the newer houdinis

 

@eetu:

Yes thats pretty much exactly what I'm looking for, You're a hero! 
The file you posted didn't work for some reason, so I had to create a new makeinsancexform and use the exact same values as you did.

Only thing I'm surprised by is that houdini does not provide an in-built solution for this, considering that normal maps are a pretty common thing to use 

Problem solved.   
Thanks a lot!

[edward]

And, by the way, the SideFX documentation on this is plain wrong: "Fit range from -1:1 to 0:1" should obviously be the other way round..

 

I wonder if what ZBrush output can depend on the various settings? If the ZBrush output are 0-1 values, then the fit from -1:1 to 0:1 makes sense to me because we had subtracted 0.5 earlier?

[eetu]

I wonder if what ZBrush output can depend on the various settings? If the ZBrush output are 0-1 values, then the fit from -1:1 to 0:1 makes sense to me because we had subtracted 0.5 earlier?

 

Well, we're talking about the normal vector here, why would you clamp it to positive values?

 

The "subtract 0.5 and normalize" does the same thing, so actually that whole step is probably extraneous..

[Sustaxata]

Sorry if thats really noobish of me, but wouldn't the substraction of 0.5 on every channel distort the normals?

[eetu]

Sorry if thats really noobish of me, but wouldn't the substraction of 0.5 on every channel distort the normals?

 

I might be wrong as I do not have ZBrush here, but a step like that is needed if the normal map is in a non-float format, like that face map above.

Normals live in the [-1..1] range, but integer file formats do not tend to store negative values. In that case the normals tend to be fitted from [-1..1] to [0..1] for saving into the image map, and then again expanded from [0..1] to [-1..1] on import to the shader. That can be node with a fit node, or by subtracting 0.5 and multiplying by to, ot normalizing instead of the multiply if you wat to be sure that the end result is normalized.

 

If the normals are brought in as .exr, then a step like that would of course distort the normals.

[freaq]

basically if it is an 8 bit map  (256 values per channel) it is like this:
0-127 the value is negative

128 no change (effectively the 0 point)

129-256 the value is positive

 

if you are mapping from 0-1:

0-5 negative

0.5-1 positive

 

to map this to floating values for normals:
(val*2)-1

 

if 8bit style index to map this to floating values for normals:

(val/128)-1

Edited March 31, 2014 by freaq

[eetu]

I tried my scene (above) at work, and it did not work right, seems like the normal map axes here are defined differently. So prepare mentally for some flipping, maybe

[freaq]

normalmaps sometimes come in "maya" encoded or "max" encoded,
basically it meas the Y and Z axis may be interchanged or flipped.
not sure if this is the ace but might be worth trying.

[acorreia]

Sorry for diging this post from the grave.

I have used the method eetu created to render my maps created in zBrush, but I have one small problem. I want to render out my images using PBR and with eetu's file unless its micropolygon renderer it comes up as black. Any solution? I have been trying but no luck whatsoever.

Cheers,
A

[dszs]

Could it be it does not have a BSDF output in the shader. PBR won't work with it. Correct me if I am wrong

[acorreia]

Yes dszs. I believe I managed to solve it, just forgot to post it here though.

Created another lighting model and conected the yellow output (BSDF as you said) and voilá, it works!

Thanks

[el_diablo]

I wonder if this is still problematic in the recent builds, or am doing something wrong? I get some problems on 8-bit normal maps, withouth the eetu's math, which look like clamping of values even if I choose the -1,1 on normal map options in the mantra surface. With the fix it works out ok.

[el_diablo]

Anyone know whats the expected output of 'Make Instance Transform' with N,v,up connected as in eetu's Normal map example. I can't seem to find any extended node documentation, as to how the 4x4 matrix output is calculated from the numerous inputs.

 

I must be missing something simple regarding the operation of that node. I understand how transform matrices are usually composed, but would love for some info on that specific vop node.

[eetu]

Have you checked the illuminating example image in bottom of http://www.sidefx.com/docs/houdini13.0/copy/instanceattrs (from wolfwood iirc)