Issues with Render to Texture

[MarkusMiller]

Hello,

I've got problems getting a render to texture scene to render. So far I have an osgViewer with a default camera. If I attach my root node with the glsl_simple.osgt scene to the osgViewer, everything renders just fine.

I then tried to add a render to texture camera to the root node, attach the scene to be rendered to it and add a canvas geometry to the root node. This is the code I use to generate the camera:

osg::Camera* setupRTTCamera(float minX, float maxX, float minY, float maxY, const osg::ref_ptr<osg::Texture2D>& tex)
{
    osg::ref_ptr<osg::Camera> cam = new osg::Camera;
    cam->setClearColor(osg::Vec4(1.f, .0f, .0f, 1.f));
    cam->setClearMask(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
    cam->setViewport(0, 0, 1280, 720);

//    cam->setProjectionMatrixAsOrtho2D(minX, maxX, minY, maxY);
//    cam->setProjectionMatrixAsOrtho(minX, maxX, minY, maxY, -5.f, 5.f);
    cam->setProjectionMatrix(osgViewer->getCamera()->getProjectionMatrix());
    cam->setViewMatrix(osgViewer->getCamera()->getViewMatrix());

    cam->setRenderOrder(osg::Camera::PRE_RENDER);
    cam->setReferenceFrame(osg::Camera::RELATIVE_RF);
    cam->setRenderTargetImplementation(osg::Camera::FRAME_BUFFER_OBJECT);
    cam->setName("RTTCamera");

    cam->attach(osg::Camera::COLOR_BUFFER, tex);

    return cam.release();
}

This is the code I use to create the canvas geometry:

osg::Geode* devlab::MockSceneManager::setupCanvas(float minX, float maxX, float minY, float maxY, const osg::ref_ptr<osg::Texture2D>& tex)
{
    osg::ref_ptr<osg::Geometry> canvasGeometry = new osg::Geometry();

    osg::ref_ptr<osg::Vec3Array> vertexArray = new osg::Vec3Array;
    vertexArray->push_back(osg::Vec3(minX, .0f, minY));
    vertexArray->push_back(osg::Vec3(maxX, .0f, minY));
    vertexArray->push_back(osg::Vec3(minX, .0f, maxY));
    vertexArray->push_back(osg::Vec3(maxX, .0f, maxY));
    canvasGeometry->setVertexArray(vertexArray);

    osg::ref_ptr<osg::Vec4Array> colourArray = new osg::Vec4Array;
    colourArray->push_back(osg::Vec4(1.f, 1.f, 1.f, 1.f));
    canvasGeometry->setColorArray(colourArray, osg::Array::BIND_OVERALL);

    osg::ref_ptr<osg::Vec3Array> normalArray = new osg::Vec3Array;
    normalArray->push_back(osg::Vec3(0.f, -1.f, 0.f));
    canvasGeometry->setNormalArray(normalArray, osg::Array::BIND_OVERALL);

    osg::ref_ptr<osg::Vec2Array> texCoordArray = new osg::Vec2Array;
    texCoordArray->push_back(osg::Vec2(0.f, 0.f));
    texCoordArray->push_back(osg::Vec2(1.f, 0.f));
    texCoordArray->push_back(osg::Vec2(0.f, 1.f));
    texCoordArray->push_back(osg::Vec2(1.f, 1.f));
    canvasGeometry->setTexCoordArray(0, texCoordArray);

    canvasGeometry->addPrimitiveSet(new osg::DrawArrays(osg::PrimitiveSet::TRIANGLE_STRIP, 0, 4));

    canvasGeometry->getOrCreateStateSet()->setTextureAttributeAndModes(0, tex, osg::StateAttribute::ON);

    osg::ref_ptr<osg::Geode> canvasNode = new osg::Geode();
    canvasNode->addDrawable(canvasGeometry);

    return canvasNode.release();
}

And I use this function to set the entire scene up:

void loadGeometry()
{
    osg::ref_ptr<osg::ClearNode> clearNode(new osg::ClearNode);
    clearNode->setClearColor(osg::Vec4(.392f, .584f, .929f, 1.f));

    root->addChild(clearNode);

    osg::ref_ptr<osg::Node> fileNode = osgDB::readNodeFile(std::string("../../data/glsl_simple.osgt"));

    transform = new osg::MatrixTransform;
    transform->setMatrix(osg::Matrix::rotate(osg::DegreesToRadians(60.), osg::Vec3(-1.f, 0.f, 0.f)) * osg::Matrix::translate(-2.f, 0.f, -20.f));

    transform->addChild(fileNode.get());


    osg::ref_ptr<osg::Texture2D> tex = new osg::Texture2D();
    tex->setTextureSize(1280, 720);
    tex->setInternalFormat(GL_RGBA);
    tex->setFilter(osg::Texture2D::MIN_FILTER, osg::Texture2D::LINEAR);
    tex->setFilter(osg::Texture2D::MAG_FILTER, osg::Texture2D::LINEAR);

    osg::ref_ptr<osg::Camera> rttCam = setupRTTCamera(-1.f, 1.f, -1.f, 1.f, tex);
    osg::ref_ptr<osg::Geode> rttCanvas = setupCanvas(-1.f, 1.f, -1.f, 1.f, tex);

    rttCam->addChild(transform);

    root->addChild(rttCam);
    root->addChild(rttCanvas);
}

And eventually, I configure the osgViewer like this in the constructor of the app:

osgViewer->setUpViewInWindow(500, 100, 1280, 720);
osgViewer->setThreadingModel(osgViewer::Viewer::SingleThreaded);

loadGeometry();

osgViewer->getCamera()->setProjectionMatrixAsOrtho(-2.f, 2.f, -2.f, 2.f, -2.f, 2.f);
osgViewer->getCamera()->setViewMatrixAsLookAt(osg::Vec3(0.f, -1.f, 0.f), osg::Vec3(0.f, 0.f, 0.f), osg::Vec3(0.f, 0.f, 1.f));

With this, I can actually see the canvas, which, however, is coloured plain red (the clear colour of the rtt-camera). The animated scene, however isn't visible on the canvas geometry.

What did I do wrong? I tried to stick to the osgprerender and osgdistortion examples, but they are unfortunately a rather maximum minimum example.

I further think that I haven't understood how cameras in OSG work, because I first tried to set the rtt-camera as a normal camera that should display the canvas (in plain white). So I just set the camera up like this:

osg::Camera* setupRTTCamera(float minX, float maxX, float minY, float maxY, const osg::ref_ptr<osg::Texture2D>& tex)
{
    osg::ref_ptr<osg::Camera> cam = new osg::Camera;
    cam->setClearColor(osg::Vec4(1.f, .0f, .0f, 1.f));
    cam->setClearMask(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
    cam->setViewport(0, 0, 1280, 720);

    cam->setProjectionMatrixAsOrtho(minX, maxX, minY, maxY, -5.f, 5.f);
    cam->setViewMatrixAsLookAt(osg::Vec3(0.f, -1.f, 0.f), osg::Vec3(0.f, 0.f, 0.f), osg::Vec3(0.f, 0.f, 1.f));

    return cam.release();
}

and added the rttCanvas as a Child. All I could see was the red clear colour, but no white geometry. How does OSG behave if there are cameras present in a subgraph?

Could you please give me a hint?

Thank you

[MarkusMiller]

Damn it... I've got the render to texture working by chaning
cam->setReferenceFrame(osg::Camera::RELATIVE_RF);
to
cam->setReferenceFrame(osg::Camera::ABSOLUTE_RF);

But I'd still be interested if my mental model is wrong or not regarding "chained cameras" in a graph.

Thank you