Reconstruction from Depth, correct value to reconstruct

simotix · 2011-06-19T13:23:33

I have been attempting to correctly reconstruct the depth buffer, and I have been able to reconstruct the value. However, I believe I am reconstructing the wrong value. I have been following http://mynameismjp.wordpress.com/2010/09/05/position-from-depth-3/ but the terminology is not always clear. Something that I noticed is that the value I have in my working model of Deferred Rendering is different then what I have in this version (for the position). I debugged the same pixel and even though the value of positionVS from "length(positionVS)" in my rendering of the buffer is the same as the value in my lighting shader, this "positionVS" is different then if I were to generate the position like this (traditional) float depthVal = depthMap.Sample( sampPointClamp, texCoord ).r; float4 position; position.xy = input.ScreenPosition.xy; position.z = depthVal; position.w = 1.0f; position = mul(position, InvertViewProjection); position /= position.w; This leads me to believe that maybe I am calculating positionVS from my vertex shader in my gbuffer wrong. As an example, in my rendering of the GBuffer my position will be "input.VertexShaderOutput::PositionVS ( -0.496, -1.000, 1.496, 1.000 ) ". In the traditional way, my position will be, "( -0.496, -0.000, -0.504, 1.000 ) " where as in this reconstruction method, it will be exactly " ( -0.496, -1.000, 1.496, 1.000 )". There is my shader code, could someone please take a look at my terminology and make sure I am calculating the correct thing? I am transposing my world, view and proj matrix before I send them into my shader. G Buffer struct VertexShaderInput { float4 Position : POSITION; float3 Normal : NORMAL; float2 TexCoord : TEXCOORD0; }; struct VertexShaderOutput { float4 PositionCS : SV_POSITION; float4 PositionVS : Position; float2 TexCoord : TEXCOORD0; float3 Normal : TEXCOORD1; }; VertexShaderOutput VertexShaderFunction(VertexShaderInput input) { input.Position.w = 1.0f; VertexShaderOutput output = (VertexShaderOutput)0; output.PositionCS = mul( input.Position, World ); output.PositionCS = mul( output.PositionCS, View ); output.PositionVS = output.PositionCS; output.PositionCS = mul( output.PositionCS, Projection ); output.TexCoord = input.TexCoord; output.Normal = mul( input.Normal, (float3x3)World ); output.Normal = normalize(output.Normal); return output; } struct PixelShaderOutput { float4 Color : SV_Target0; float4 Normal : SV_Target1; float4 Depth : SV_Target2; }; PixelShaderOutput PixelShaderFunction(VertexShaderOutput input) { PixelShaderOutput output = (PixelShaderOutput)0; output.Color = txDiffuse.Sample( samLinear, input.TexCoord ); output.Color.a = SpecularIntensity; output.Normal.rgb = 0.5f * (normalize(input.Normal) + 1.0f); output.Normal.a = SpecularPower; output.Depth.x = length(input.PositionVS); return output; }; Point Light: struct VertexShaderInput { float3 Position : POSITION0; }; struct VertexShaderOutput { float4 PositionCS : SV_POSITION; float3 ViewRay : TEXCOORD0; }; VertexShaderOutput VertexShaderFunction(VertexShaderInput input) { VertexShaderOutput output; float4 inputPosition = float4(input.Position, 1.0f); output.PositionCS = mul( inputPosition, World ); float3 positionWS = output.PositionCS.xyz; output.ViewRay = positionWS - CameraPosition; output.PositionCS = mul( output.PositionCS, View ); output.PositionCS = mul( output.PositionCS, Projection ); return output; } float4 PixelShaderFunction(VertexShaderOutput input) : SV_Target { float2 texCoord = input.PositionCS.xy * InvTextureSize; float3 viewRay = normalize(input.ViewRay); float viewDistance = depthMap.Sample( sampPointClamp, texCoord ).x; float3 positionWS = CameraPosition + viewRay * viewDistance; float4 normalData = normalMap.Sample( sampPointClamp, texCoord); float3 normal = 2.0f * normalData.xyz - 1.0f; float3 lightDir = normalize(positionWS - PointLightPosition); float d = length( PointLightPosition - positionWS); float nDl = dot(normal, lightDir); float atten = 1.0f -d/PointLightRadius; float4 color = (atten * float4(PointLightColor.rgb, 1.0f)) + (nDl * atten); return color; }

Graphics and GPU Programming Programming

Started by simotix February 13, 2011 07:57 PM

45 comments, last by csisy 12 years, 10 months ago

simotix

142

Author

March 21, 2011 04:22 AM

You said something about 'viewray should be multipled by the word matrix', so i thought that that would be the most appropriate answer.
Didn't really know what you were getting at to be honest.

The way you you and MJP calculate your eyeRay/eyeToPixel is considerably different, I was wondering why you do your the way that you do?

here's my pointlight shader using MJPs method for reconstruction

VS_OUTPUT_LIGHTPASS_INSTANCE vs_lightPass(VS_INPUT_INSTANCE In) { In.Pos.xyz *= In.insPos.w; In.Pos.xyz += In.insPos.xyz; Out.vEyeRay = In.Pos.xyz - camPos.xyz; } PS_output ps_pointLight(in VS_OUTPUT_LIGHTPASS_INSTANCE In ) { const float3 eyeToPixel = normalize(In.vEyeRay.xyz); }

MJPS



VS_OUTPUT VS( VS_INPUT Input )

{

	float4 positionScaled = float4(Input.PositionOS * Scale, 1.0f);

	Output.PositionWS = mul(positionScaled, World).xyz;

	Output.ViewRay = Output.PositionWS - CameraPosWS;



	return Output;

}



PS_OUTPUT_RECONSTRUCT PSReconstructLinear(PS_INPUT Input)

{   	

	float3 viewRay = normalize(Input.ViewRay);

	float3 positionWS = CameraPosWS + viewRay * depth;

	return output;

}

So have you found your error?

Not yet unfortunately

simotix

142

Author

March 23, 2011 04:56 AM

Are there any other suggestions?

simotix

142

Author

March 25, 2011 03:44 AM

Can anyone provide any suggestions?

Daniel E

224

March 25, 2011 04:26 AM

The way you you and MJP calculate your eyeRay/eyeToPixel is considerably different, I was wondering why you do your the way that you do?

It isn't really different, I'm just rendering the point lights with hardware instancing and scale and offset the vertices with the instance data (In.insPos).

I'm successfully using his method for point lights and directional lights.

Unfortunately I have no new suggestions, but I could fire up rendermonkey real quick and try to make a minimal example.

[size="1"]sourceforge.net/projects/megatexturedemo

Daniel E

224

March 25, 2011 05:50 AM

Alright so here is a point light example for RenderMonkey:

http://www.mediafire.com/file/iy62h4jahm5pgcd/pointlight.zip

Hope it helps.

Edit: aligned texels to pixels, now it's even looking correct

[size="1"]sourceforge.net/projects/megatexturedemo

csisy

400

June 16, 2011 08:58 AM

Hi

I have a deferred lighting system. It works correctly if the Depth is in a r32f rendertarget, and the depth = Position.z / position.w (where Position is the transformed object space position with view and projection)
But I'd like to use a linear depth with point lights (with bounding volumes = spheres)

Here my code: (it has some hacks which will be removed when it will work

)

GBUFFER

//vertex shader



float4x4 world;

//float4x4 viewProj;

float4x4 view;

float4x4 proj;



void main(

	in	out	float4 Position	: POSITION0,

	in		float3 Normal	: NORMAL0,

	in	out	float2 Texcoord	: TEXCOORD0,

	in		float3 Tangent	: TANGENT0,

	in		float3 Binormal	: BINORMAL0,



		out	float2 Depth	: TEXCOORD1,

		out float3x3 Wtt	: TEXCOORD2)

{

	float4 wpos = mul(Position, world);

	float4 vpos = mul(wpos, view);

	Position = mul(vpos, proj);

	//Position = mul(wpos, viewProj);



	Depth.x = vpos.z;

	Depth.y = 30.0f;

	//Depth.x = Position.z;

	//Depth.y = Position.w;



	Wtt[0] = mul(normalize(Tangent), world);

	Wtt[1] = mul(normalize(Binormal), world);

	Wtt[2] = mul(normalize(Normal), world);

}



//pixel shader



sampler2D samplerDiffuse : register(s0);

sampler2D samplerNormal  : register(s1);



void main(

	in		float2 Texcoord	: TEXCOORD0,

	in		float2 Depth	: TEXCOORD1,

	in		float3x3 Wtt	: TEXCOORD2,



		out	float4 RT0		: COLOR0,

		out	float4 RT1		: COLOR1,

		out	float4 RT2		: COLOR2)

{

	Texcoord *= 5.0f;

	RT0.xyz = tex2D(samplerDiffuse, Texcoord).xyz;

	RT0.w = 1.0f;



	float3 normal = 2.0f * tex2D(samplerNormal, Texcoord) - 1.0f;

	normal = mul(normal, Wtt);

	RT1.xyz = 0.5f * (normalize(normal) + 1.0f);

	RT1.w = 1.0f;



	RT2 = Depth.x / Depth.y;

}

And the point light shader

//vertex shader



float4x4 world;

float4x4 viewProj;



void main(

	in	out	float4 Position		: POSITION0,

		out float4 ScreenPos	: TEXCOORD0,

		out float3 WorldPos		: TEXCOORD1)

{

	float4 wpos = mul(Position, world);

	Position = mul(wpos, viewProj);



	ScreenPos = Position;

	WorldPos = wpos.xyz;

}



//pixel shader



//....



float3 viewRay = normalize(WorldPos - cameraPosition);

float3 position = cameraPosition + viewRay * depthText;

//...

The position is incorrect, so the lighting is also wrong. I'd be happy if someone could help me. ;)

sorry for my bad english

csisy

400

June 18, 2011 07:19 AM

Can anyone help me?

sorry for my bad english

csisy

400

June 19, 2011 01:23 PM

[color=#1C2837][size=2]Can anyone help me?

sorry for my bad english

Reconstruction from Depth, correct value to reconstruct

This topic is closed to new replies.

Popular Topics

Recommended Tutorials

Reconstruction from Depth, correct value to reconstruct

This topic is closed to new replies.

Popular Topics

Recommended Tutorials

Reticulating splines