RE: tangent maps

["Halfdan Ingvarsson" <hingvars(at)Softimage.COM>]

Title: Re: tangent maps

WARNING: This e-mail may cause nausea, headaches and/or breathing difficulties.

 

If I'm not mistaken, it gives you the dot product between the calculated/smoothed tangent (T) and the cross product of the calculated/smoothed binormal (B) and the vertex normal (N). To wit:  d = - T . ( B x N )

 

Since the binormal and the tangent are calculated mostly independently, you may end up with a left-handed coordinate system, rather than a right-handed one, depending on the underlying texture projection that they're calculated from. This value will then tell you whether to do reverse the cross product of the tangent and the normal when calculating the binormal to end up with a proper right-handed coordinate system.

 

Most of the time it probably doesn't matter but it's there in case you need it.

 

 - ½

---

From: owner-xsi(at)Softimage.COM on behalf of André Adam
Sent: Fri 26-Jan-07 03:58
To: XSI(at)Softimage.COM
Subject: Re: tangent maps

I've a whole bunch of programmers around here that got headaches because
of those alpha values; actually we did not find out up to now and my
personal opinion still is it simply is trash (though I'm the only one
over here with that opinion...). Anyone from Soft like to give us a hint?
That only 8bit data thing for the tangents is odd, since they use float
arrays... on the other hand, that whole normal mapping technique is
unprecise as hell anyways, so I guess nobody will ever notice... ;)


Ben Rogall wrote:
>
>
> Thanks for forwarding that Andre. The hardest part about the vector
> displacement was setting up the tangent space conversion. Well, that
> and trying different ways of exporting objects and mapping the
> position data. Mental Ray already knows how to do multi-directional
> displacement.
> Regarding the bit-depth of the tangent property, if I read out any of
> the values, say by checking the SDK explorer and multiply by 255, I
> get a whole number. Btw, I have no idea what is being stored in the
> alpha channel.
>
>
>
> --- On Thu 01/25, =?ISO-8859-1?Q?Andr=E9_Adam?= < a_adam(at)49games.de >
> wrote:
>
>     *From: *=?ISO-8859-1?Q?Andr=E9_Adam?= [mailto: a_adam(at)49games.de]
>     *To: *XSI(at)Softimage.COM
>     *Date: *Thu, 25 Jan 2007 17:32:02 +0100
>     *Subject: *Re: tangent maps
>
>     Hi Ben,
>
>     thanks for the kind words! Very nice work on the displacement mapping!
>     A vertex colour property stores float values, not 8bit data, so
>     you're
>     practically lossless when converting back to vectors. The smoothing
>     works similar, since it is angle based, but is not connected in
>     any way
>     to the geo approximation or the normals. In theory, the normal map
>     should be able to compensate cracks in the tangent space, but
>     practically the normal map's texels would have to line up
>     perfectly with
>     the polygon edges which is a very unlikely occasion. So, as soon
>     as you
>     try to fit a normal map texture with its pixel raster onto uneven 3d
>     geometry, the tangent space needs to be smoothed to avoid cracks
>     and seams.
>
>     I've attached the script to this mails, but really, this is nothing
>     fancy, just a one-time creation of null objects pointing in the
>     vector's
>     direction. A compiled operator with live-input certainly would be way
>     nicer. Oh, and it is slow, be careful with meshes heavier than -
>     say - a
>     thousand polygons, trying it out on small objects will be more
>     fun... :)
>
>     Cheers!
>
>     -André
>
>
>     Ben Rogall wrote:
>     >
>     > Andre, very very nice explanation. I've been using XSI tangents
>     > recently in experiments with tangent space vector displacement:
>     >
>     http://www.xsibase.com/forum/index.php?board=29;action="">
>     > It took me a little while to realize I had to do the [0,1] to
>     [-1,1]
>     > interval correction. I'm also curious to see whether having only 8
>     > bits per channel for the tangent vector is enough to avoid jumping
>     > though I haven't seen anything obvious yet.
>     >
>     > Another issue with tangent space is difficulty when you try to get
>     > different 3D applications to work with eachother. Once you have the
>     > normal and tangent vectors, there are still several ways to
>     determine
>     > your T,B,N space depending on which of these vectors you decide to
>     > orthoganalize to the others. This choice has to match between the
>     > applications.
>     >
>     > I hadn't even thought about the smoothing issue you mentioned.
>     Is that
>     > analogous to the normal automatic discontinuity control?
>     >
>     > Also is your tangent display utility publicly available?
>     >
>     > Thanks,
>     > Ben
>     >
>     >
>     > --- On Thu 01/25, =?ISO-8859-1?Q?Andr=E9_Adam?= <
>     a_adam(at)49games.de >
>     > wrote:
>     >
>     > *From: **=?ISO-8859-1?Q?Andr=E9_Adam?= [mailto: a_adam(at)49games.de]
>     > *To: *XSI(at)Softimage.COM
>     > *Date: *Thu, 25 Jan 2007 11:20:09 +0100
>     > *Subject: *Re: tangent maps
>     >
>     > Hi Bernard,
>     >
>     > the tangent property in XSI is the so called tangent map, or tangent
>     > space. Encoded in there you can find a given texture space's
>     > u-direction
>     > (the reason why you have to reference a texture projection when
>     > creating
>     > a tangent property) in 3d space (you could call this the "texture
>     > mapped" u-direction) on a per sample basis, so we're talking of
>     > vectors
>     > here. In absence of a more convenient property Softimage decided to
>     > encode the tangents into vertex colors. That means you have a
>     > little bit
>     > of math to do to revert them back to vector data: (VC-0.5)*2. This
>     > formula stretches the RGB vertex color values, which have a 0-1
>     > range,
>     > back to a vector's -1 to +1 range.
>     >
>     > The reason you need a tangent space when dealing with normal maps is
>     > that a "normal map in tangent space" (there is also an object
>     > space flavour)
>     > stores relative information as to where a given normal should be
>     > bent,
>     > which after all is what bump mapping is all about (bump mapping and
>     > normal mappinng are practically the same thing). The lila-blueish
>     > color
>     > in a tangent space normal map means that the geometry's input normal
>     > should not be altered, where cyan, magenta, etc colours indicated a
>     > shift in normal orientation.
>     > This shift, which is a relative bit of information as it does not
>     > point
>     > to global world coordinates, needs a reference on the object
>     > itself from
>     > which the relative shift should be calculated - the tangent space.
>     > The
>     > tangent space offers a predefined vector for each sample point
>     and is
>     > directly generated from the texture projection the normal map is
>     > mapped
>     > with onto the object.
>     >
>     > I've written a small utility that creates little pointers from the
>     > tangent space right on the object's surface to better show the
>     > vectors
>     > encoded there. I've attached two screenshots which might clearify
>     > that a
>     > bit. The first one shows a straight spherical mapping, the second
>     > one is
>     > twisted. As soon as a texture space shows distortions (or even
>     worse:
>     > seams), just like the second one, tangent spaces become tricky; the
>     > samples for a single vertex point in different directions, which
>     > finally
>     > would create artifacts during normal mapping. To cope with this
>     issue
>     > the tangent operator in XSI has a smoothing value, which basically
>     > middles the different sample's tangent vectors using an angle-based
>     > approach.
>     >
>     > Since the normal map stores its data relative to the tangent space,
>     > generation, as done with the ultimapper for example, therefore
>     always
>     > needs to reference it during generation to provide correct results.
>     > Ultimapper basically marries referenced object curvature from a high
>     > resolution source object with the normals of the low resolution
>     > source
>     > object and the tangent space of the low resolution object.
>     >
>     > The Photoshop plugins and tools provided by graphics hardware
>     vendors
>     > don't have this kind of 3d information available when generating
>     > normal
>     > maps from heightmaps. They assume that the object the generated
>     > normal
>     > map might get mapped onto shows an ideal tangent space, which
>     > usually is
>     > nonsense. The more distorted the texture space on the mapped
>     > object is,
>     > the wronger (is that english?!) the bump mapping results with such a
>     > textures are. I'd stay away from these tools for production work.
>     >
>     > Hope that makes some sense, cheers!
>     >
>     > -André
>     >
>     /*========================================================
>     "Show Tangents" / André Adam
>     Tool to create null object pointers on an polymesh object's surface
>     to visualise direction in a tangent space.
>     ========================================================*/
>
>
>     function fGetPointer(oObj, oSize){
>     var oNull, oDisplay
>     >>     oNull.Parameters("Primary_Icon").Value = 9;
>     oNull.Parameters("Size").Value = oSize;
>     oDisplay = oNull.AddProperty("Display Property");
>     oDisplay.Parameters("wirecol").Value = Math.round(Math.random()*1023);
>     return oNull;
>     }
>
>     function fGetTangents(oPolyMsh){
>     var eVCs = new Enumerator(oPolyMsh.VertexColors);
>     while(!eVCs.atEnd()){
>     if(eVCs.item().Name == "Tangents"){
>     return new VBArray(eVCs.item().Elements.Array).toArray();;
>     }
>     eVCs.moveNext();
>     }
>     return false;
>     }
>
>     function fGetSamplePositions(oPolyMsh){
>     var aSamplePositions = new Array();
>     var ePoints = new Enumerator(oPolyMsh.Vertices);
>     for(var i=0; i
>     ePoints.moveFirst();
>     while(!ePoints.atEnd()){
>     var eSamples = new Enumerator(ePoints.item().Samples);
>     while(!eSamples.atEnd()){
>     if(eSamples.item().Index == i){
>     aSamplePositions.push(ePoints.item().Position);
>     }
>     eSamples.moveNext();
>     }
>     ePoints.moveNext();
>     }
>     }
>     return aSamplePositions;
>     }
>
>     function fShowTangents(oObj, oSize){
>     var oPolyMsh = oObj.ActivePrimitive.Geometry;
>     var aTangents = fGetTangents(oPolyMsh);
>     if(aTangents){
>     var aSamplePositions = fGetSamplePositions(oPolyMsh);
>     for(var i=0; i var vY =
>     XSIMath.CreateVector3((aTangents[i]-0.5)*2,
>     (aTangents[i+1]-0.5)*2, (aTangents[i+2]-0.5)*2);
>     var vX = XSIMath.CreateVector3(0, 1, 0);
>     var vZ = XSIMath.CreateVector3();
>     vZ.Cross(vY, vX);
>     vX.Cross(vY, vZ);
>     vX.Normalize(vX);
>     vY.Normalize(vY);
>     vZ.Normalize(vZ);
>     var oM3 = XSIMath.CreateMatrix3(vX.X, vX.Y, vX.Z, vY.X, vY.Y,
>     vY.Z, vZ.X, vZ.Y, vZ.Z);
>     var oTrans = XSIMath.CreateTransform();
>     oTrans.SetRotationFromMatrix3(oM3);
>     oTrans.SetTranslation(aSamplePositions[i/4]);
>     oTrans.SetScalingFromValues(0.25, 1, 0.25);
>     var oPointer = fGetPointer(oObj, oSize);
>     oPointer.Kinematics.Local.Transform = oTrans;
>     }
>     }
>     else{
>     LogMessage("aa_ShowTangents: No tangents found on object!", siError);
>     }
>     }
>
>     /*========================================================
>     MAIN
>     ========================================================*/
>
>     if((Selection.Filter.Name == "object") && (Selection.Count == 1)
>     && (Selection(0).Type == "polymsh")){
>     fShowTangents(Selection(0), 1);
>     }
>     else{
>     LogMessage("aa_ShowTangents: Not a single polymesh object
>     selection!", siError);
>     }
>
> ------------------------------------------------------------------------
> *Join Excite! - http://www.excite.com*
> The most personalized portal on the Web!

---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
unsubscribe xsi