Parallax mapping troubles :(

[M/\dm/\n]

I've been experimenting with parallax, maping yesterday, but faced some problems.

What I've been doing was:
1st: gluLookAt
2nd: draw geometry stage + passing tangent, binormal, normal to vp
3rd: passing eyepos, + MVP,MV,MVIT matrices to cg vp
4th: cg vp -> mul(mvit,eyepos), E=eyepos-IN.pos, mul(TBN, E)
5th: cg fp, fetch height from texture, normalize E, use the function to calculate offset offset=(h*E.xy/E.z)
The problem seems to be connected with transformations, as the texture appears to rotate vhen crossing some point (moving with gluLookAt)

[dorbie]

Moving to the Advanced forum.

[Zengar]

Is MVIT the inverse-transpose? I thought it should be simply inverce. But MVI*eye is always (0, 0, 0). I don't know if this can help, but...

[M/\dm/\n]

Whoups, yes you're right must start coding more carefully.

Now it's only sharp angle artifacts to get rid of
I'll have to sit a while to translate fp code to Cg, unfortunately I mised good ol' assembly But docs rely on it...

Hope vp is correct now:

struct a2v
{
    float4 Position		: POSITION;
    float2 TexCoord0	: TEXCOORD0;
};
 
struct v2f
{
    float4 HPosition	: POSITION;
	float2 TexCoord0	: TEXCOORD0;
	float3 Eye			: TEXCOORD1;
};
 
v2f main(a2v IN,
		 uniform float4x4 ModelViewProj,
		 uniform float4 EyePos,
		 uniform float4 T,
		 uniform float4 B,
		 uniform float4 N
		 )
{
    v2f OUT;
 
	OUT.HPosition = mul(ModelViewProj, IN.Position);
	OUT.TexCoord0 = IN.TexCoord0;
 
	float3x3 TBN = float3x3(T.xyz, B.xyz, N.xyz);
	float3 E = EyePos.xyz-IN.Position.xyz;
	E = mul(TBN, E);
 
	OUT.Eye = E;
 
    return OUT;
}

[This message has been edited by M/\dm/\n (edited 02-09-2004).]

[Zengar]

Well, I also wrote a parallax mapping app (about 5 minutes ago) and so far i can see, my vp looks exact ys yours one.
Are you drawing some sort of patch? This would explain your static TBN vectors

[M/\dm/\n]

Yup, right now I'm trying to sort things out on plane, then I'll go into 3D + I think I'll pass only 2 of 3 TBN vectors to GPU, 3rd will be cross on GPU (gonna see how FX5200 will handle that). And I've learned that uniform parameters are real pain in a** when you have 10+ shaders in one app.

[This message has been edited by M/\dm/\n (edited 02-09-2004).]

[M/\dm/\n]

Huh, I think I've finally got algo working

But I noticed that there are prety ugly artifacts when parallax is applied on some of my textures:
W/O parallax
http://www.lapas.dau.lv/salitis/1.bmp
W parallax
http://www.lapas.dau.lv/salitis/2.bmp
Tried experimenting with biases, but that doesn't seem to be a good solution, as you'll have flaws or no parallax
Of course my textures arn't very nice as they have a lot of sharp angles.

BTW, cg fp:

void main(float2 TexCoord0	: TEXCOORD0,
	  float3 Eye		: TEXCOORD1,
 
	  out float4 color : COLOR,
 
	  uniform sampler2D   diffuseMap,
	  uniform sampler2D   heightMap
	  )
{
	normalize(Eye);
	float3 height = tex2D(heightMap, TexCoord0).xyz*0.04-0.02;
 
	float4 tmpcol = tex2D(diffuseMap, TexCoord0+(height*Eye.yx));
 
	float3 C = tmpcol.xyz*tmpcol.w;
	color.xyz = C;
	color.w = 1;
}

I'm storing lightmap in alpha of diffuse map. Still can't understand why I had to swap Eye.xy to Eye.yx.