Skip to content

GitLab

Commit eaaf7b32 authored by Sebastian Eichelbaum's avatar Sebastian Eichelbaum
Browse files

[CHANGE] some improved defaults and more documentation

parent c141e516
Hide whitespace changes
Inline Side-by-side
Showing with 50 additions and 40 deletions
src/core/graphicsEngine/postprocessing/WGEPostprocessorLineAO.cpp
View file @ eaaf7b32
......@@ -56,7 +56,7 @@ WGEPostprocessorLineAO::WGEPostprocessorLineAO( osg::ref_ptr< WGEOffscreenRender
lineaoRadiusSS->setMax( 10.0 );
WPropDouble lineaoTotalStrength = m_properties->addProperty( "Total Strength", "The strength of the effect. Higher values emphasize the effect.",
0.5 );
1.0 );
lineaoTotalStrength->setMin( 0.0 );
lineaoTotalStrength->setMax( 5.0 );
......@@ -81,7 +81,9 @@ WGEPostprocessorLineAO::WGEPostprocessorLineAO( osg::ref_ptr< WGEOffscreenRender
new WGEShaderPropertyDefine< WPropInt >( "WGE_POSTPROCESSOR_LINEAO_SAMPLES", lineaoSamples ) )
);
// create the rendering pass
// create the LineAO rendering pass
osg::ref_ptr< WGEOffscreenTexturePass > pass = offscreen->addTextureProcessingPass( s, "LineAO" );
pass->getOrCreateStateSet()->addUniform( new WGEPropertyUniform< WPropDouble >( "u_lineaoDensityWeight", lineaoDensityWeight ) );
pass->getOrCreateStateSet()->addUniform( new WGEPropertyUniform< WPropDouble >( "u_lineaoTotalStrength", lineaoTotalStrength ) );
......
src/core/graphicsEngine/shaders/shaders/WGEPostprocessor-fragment.glsl
View file @ eaaf7b32
......@@ -289,6 +289,15 @@ float getSSAO()
#ifdef WGE_POSTPROCESSOR_LINEAO
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
//
// This implements the LineAO algorithm from http://doi.ieeecomputersociety.org/10.1109/TVCG.2012.142
//
// - This implementation matches the paper in most cases. We made some additional improvements and
// simplifications here. This simply is due to the time-lag between first submission and final acceptance.
//
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
#ifndef WGE_POSTPROCESSOR_LINEAO_SAMPLES
#define WGE_POSTPROCESSOR_LINEAO_SAMPLES 32
#endif
......@@ -299,32 +308,18 @@ float getSSAO()
/**
* The total influence of SSAO.
*/
//uniform float u_lineaoTotalStrength = 2.5; // lines
uniform float u_lineaoTotalStrength = 2.5; // brain lines
//uniform float u_lineaoTotalStrength = 2.25; // brain lines video
//uniform float u_lineaoTotalStrength = 1.5; // smallfibs
//uniform float u_lineaoTotalStrength =2.0; // smallfibs-corticospinal-tubes
//uniform float u_lineaoTotalStrength =2.5; // smallfibs-corticospinal-lines
//uniform float u_lineaoTotalStrength = 2.0; // deltawing tube
//uniform float u_lineaoTotalStrength = 3.0; // deltawing lines
uniform float u_lineaoTotalStrength = 1.0; // Should be the same default as in WGEPostprocessorLineAO.cpp
/**
* The radius of the hemispshere in screen-space which gets scaled.
*/
//uniform float u_lineaoRadiusSS = 2.0; // lines
uniform float u_lineaoRadiusSS = 2.5; // brain lines
//uniform float u_lineaoRadiusSS = 2.0; // brain lines-video
//uniform float u_lineaoRadiusSS = 2.5; // smallfibs
//uniform float u_lineaoRadiusSS = 2.5; // brain lines -corticospinal
//uniform float u_lineaoRadiusSS = 1.0; // deltawing tube
//uniform float u_lineaoRadiusSS = 2.5; // deltawing lines
uniform float u_lineaoRadiusSS = 2.0; // Should be the same default as in WGEPostprocessorLineAO.cpp
/**
* The strength of the occluder influence in relation to the geometry density. The higher the value, the larger the influence. Low values remove
* the drop-shadow effect.
*/
//uniform float u_lineaoDensityWeight = 2.0; // video
uniform float u_lineaoDensityWeight = 1.0;
uniform float u_lineaoDensityWeight = 1.0; // Should be the same default as in WGEPostprocessorLineAO.cpp
/**
* Calculate the screen-space ambient occlusion LineAO for the given pixel.
......@@ -379,11 +374,15 @@ float getLineAO( vec2 where )
// this allows an adaptive radius
// NOTE: we do not exactly use the linear scaling from the paper here. Although they look very similar, the non-linear radius, which
// increases non-linearly with the level l, improves visual quality a bit.
// increases non-linearly with the level l, improves visual quality a bit. The drawback of this method is that increasing SCALERS causes
// larger structures to become more important than local structures.
#define radScaleMin 1.5
radiusScaler += radScaleMin + l; // increment radius each time.
radiusScaler += radScaleMin + l;
// Alternatvie, linear way -- more coherent AO when changing the number of hemispheres:
// radiusScaler = radScaleMin + l; // increment radius each time.
// Get SAMPLES-times samples on the hemisphere and check for occluders
int numSamplesAdded = 0; // used to count how many samples really got added to the occlusion term
for( int i = 0; i < SAMPLES; ++i )
{
// grab a rand normal from the noise texture
......@@ -395,18 +394,21 @@ float getLineAO( vec2 where )
ray = radiusScaler * radius * hemisphereVector;
ray = sign( dot( ray, normal ) ) * ray;
// if the ray is outside the hemisphere then change direction
// get the point in texture space on the hemisphere
hemispherePoint = ray + ep;
// HACK! Somehow handle borders
//if ( ( hemispherePoint.x < 0.0 ) || ( hemispherePoint.x > 0.65 ) )
//{
// hemispherePoint.x *= -1.0;
//}
//if ( ( hemispherePoint.y < 0.0 ) || ( hemispherePoint.y > 0.575 ) )
//{
// hemispherePoint.y *= -1.0;
//}
// we need to handle the case where a hemisphere point is outside the texture space. The correct solution would be to ensure that
// no hemisphere point ever gets outside the texture space. This can be achieved by rendering the scene to a larger texture and than
// process only the required viewport in this texture. Unfortunately OpenWalnut does not provide an easy to use mechanism for this.
// So, we skip hemispherePoints outside the texture. This yields nearly no or only small artifacts at the borders if zoomed in.
if( ( hemispherePoint.x < 0.0 ) || ( hemispherePoint.x > 1.0 ) ||
( hemispherePoint.y < 0.0 ) || ( hemispherePoint.y > 1.0 )
)
{
continue;
}
// Count the samples we really use.
numSamplesAdded++;
// get the depth of the occluder fragment
occluderDepth = getDepth( hemispherePoint.xy );
......@@ -418,22 +420,20 @@ float getLineAO( vec2 where )
depthDifference = currentPixelDepth - occluderDepth;
// calculate the difference between the normals as a weight. This weight determines how much the occluder occludes the fragment
// TODO(ebaum): evaluate these alternatives: the first one is the surface direction and occluder surface direction
float pointDiffuse = max( dot( hemisphereVector, normal ), 0.0 ); // this equals the diffuse term in Phong if lightDir is the
// occluder's surface
// TODO(ebaum): check normal again with another light
vec3 t = getTangent( hemispherePoint.xy ).xyz;
vec3 newnorm = normalize( cross( normalize( cross( t, normalize( hemisphereVector ) ) ), t ) );
float occluderDiffuse = max( dot( newnorm, gl_LightSource[0].position.xyz ), 0.0 );
// TODO(ebaum): better incorporate specular reflection
// incorporate specular reflection
vec3 H = normalize( gl_LightSource[0].position.xyz + normalize( hemisphereVector ) );
float occluderSpecular = pow( max( dot( H, occluderNormal ), 0.0 ), 100.0 );
// this second is as proposed for AO, the omega (hemisphere vector) and surface normal
normalDifference = pointDiffuse * ( occluderSpecular + occluderDiffuse );
normalDifference = 1.0 - normalDifference;
normalDifference = pointDiffuse * ( occluderSpecular + occluderDiffuse );
normalDifference = 1.5 - normalDifference; // we use 2 here as occluderSpecular + occluderDiffuse might get larger than 1
// the higher the depth difference, the less it should influence the occlusion value since large space between geometry normally allows
// many light. It somehow can be described with "shadowiness". In other words, it describes the density of geometry and its influence to
......@@ -452,15 +452,23 @@ float getLineAO( vec2 where )
// the following step function ensures that negative depthDifference values get clamped to 0, meaning that this occluder should have NO
// influence on the occlusion value. A positive value (fragment is behind the occluder) increases the occlusion factor according to the
// normal weight and density weight
occlusionStep += invSamples * normalDifference * densityWeight * step( falloff, depthDifference );
occlusionStep += normalDifference * densityWeight * step( falloff, depthDifference );
}
// for this radius, add to total occlusion
occlusion += occlusionStep;
// for this radius, add to total occlusion. Keep in mind to normalize the occlusion term according to the number of samples taken
occlusion += ( 1.0 / float( numSamplesAdded ) ) * occlusionStep;
}
// we need to take care when the number of hemispheres is increased. This is done indirectly by calculation of the densityInfluence variable.
// But we need some additional scaling to keep the overall image intensity (overall energy) when changing the numbr of hemispheres.
// This usually would be done by using 1/SCALERS.
float occlusionScalerFactor = 1.0 / ( SCALERS );
// also allow the user to modify the strength if he likes
occlusionScalerFactor *= u_lineaoTotalStrength;
// output the result
return clamp( ( 1.0 - ( u_lineaoTotalStrength * occlusion ) ), 0, 1 );
return clamp( ( 1.0 - ( occlusionScalerFactor * occlusion ) ), 0, 1 );
}
/**
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment