thanks for trying to shed some light on this. As I said, I have already installed the vulkan SDK (the "official" one from LunarG as I am trying to build for windows). I have even added the include directory directly in the *.pro file, and have observed that when opening the qt3d project with QtCreator, it can correctly resolve the vulkan libraries, so I think this is not the problem. I will try your suggestion of putting the vulkan headers directly in the qt directory and see if that gets me anywhere..

Ok, I figured out how to use updateData properly. Leaving this here for posterity. For my use case, I needed to upload a grayscale 3D texture but this should be the same for any texture type: using namespace Qt3DRender; QTexture * tex = new QTexture3D; tex->setSize(vol.width(), vol.height(), vol.depth()); tex->setFormat(QTexture3D::TextureFormat::R8_UNorm); tex->setLayers(1); tex->setGenerateMipMaps(false); QTextureWrapMode wm(QTextureWrapMode::ClampToBorder); tex->setWrapMode(wm); data->setFormat(QOpenGLTexture::TextureFormat::R8_UNorm); data->setWidth(vol.width()); data->setHeight(vol.height()); data->setDepth(vol.depth()); data->setMipLevels(1); data->setLayers(1); data->setPixelFormat(QOpenGLTexture::PixelFormat::Red); data->setPixelType(QOpenGLTexture::PixelType::UInt8); QByteArray qba; qba.setRawData((const char *)vol.dataUint8(), static_cast<int>(vol.byteSize())); data->setData(qba, 1); QTextureDataUpdate update; update.setX(0); update.setY(0); update.setData(data); tex->updateData(update); Qt3DRender::QParameter p(); p.setName("volume"); // the uniform name in the shader p.setValue(QVariant::fromValue(tex)); QMaterial * mat = ...; // queried from the qml mat->addParameter(p); The Qt3D test for texture update was pretty helpful in figuring this out.

I have a similar problem. I get the following error message when loading a gltf file with the QML SkeletonLoader: [Qt][Qt3D.Renderer.Jobs][2] :0 :: Unknown buffer 0 when processing buffer view Neither self-exported files from Blender nor files downloaded from the gltf git (glTF-Sample-Models) are working. So I also need to know how to correctly export gltf or whether further adjustments in Qt are necessary in order to make it work.

Or you can fix the macdeployqt code and build it to replace your version. That way you benefit from all the file processing done during the deployment process.

Hi, I can't answer to that directly. Can you provide a minimal compilable example that shows that behaviour ?

opencv (extra characters to make response longer than 8 characters...lame!)

OK, Finally we found the solution. The tip is to create a Scene3D in the qml file. And to create the camera and ForwaqrdRenderer in the QML too (and not in the C++). Scene3D { id: scene3D anchors.fill: parent anchors.margins: 10 focus: true Root3DEntity { id : root3DEntity renderSettings : RenderSettings { id : renderSettings pickingSettings.pickMethod : pickingSettings.PrimitivePicking activeFrameGraph: ForwardRenderer { id: renderer clearColor: "#4d4d4f" camera: Camera { id: camera projectionType: CameraLens.PerspectiveProjection fieldOfView: 45 nearPlane: 0.1 farPlane: 1000.0 position: Qt.vector3d(0.0, 0.0, 2.0) upVector: Qt.vector3d(0.0, 1.0, 0.0) viewCenter: Qt.vector3d(0.0, 0.0, 0.0) } } } } }

Correction. I do not why. But my proposal no more works. But I found another way that seems to work. In fact the QDiffuseSpecularMaterial that we are supposed to use (other materials are deprecated), disable the depth buffer. So, this is a working code to force the enable (create a new effect does not work): Qt3DRender::QEffect *effect = material->effect(); if (effect) { for (Qt3DRender::QTechnique* currentTechnique : effect->techniques()) { for (Qt3DRender::QRenderPass * currentPass : currentTechnique->renderPasses()) { for (Qt3DRender::QRenderState * currentState : currentPass->renderStates()) { if (dynamic_cast<Qt3DRender::QNoDepthMask *>(currentState)) { currentPass->removeRenderState(currentState); Qt3DRender::QDepthTest* depthTest = new Qt3DRender::QDepthTest; depthTest ->setDepthFunction(Qt3DRender::QDepthTest::Less); currentPass->addRenderState(depthTest); break; } } } } } Warning: if you have fully transparent (alpha = 0) objects, the test fails. So the objects behind the invisible one are not displayed.

@Alain38-0 OK. Now I have found a way to always have a displayed text. For this, in the initial setText (when scene is not yet rendered) I'm putting all the characters that can appear in the strings that will be displayed. Then, I have something displayed. But it is often wrong. To be more explicit I'm doing this: void createScene() { ... myTextEntity->setText("?CLRT0123456789"); ... } void pageDisplayed() { ... myTextEntity->setText("???"); ... } void onButtonCLicked() { myTextEntity->setText("C23"); } setText("C23") displays another string of three characters (like "2R7"). But, at least strings have a coherency (all 'L' are replaced by '0', and so on).

@julianm You will have to set setSyncData(true); on the attribute-buffers and also synch the index buffers to get the right vertices.

Ok I found what is causing that - the torus is created facing Z. Thanks.

Great question, @GerryH! My impression is that the automotive industry is one of biggest customers for the Qt Company (TQtC) at the moment. 3D graphics and Qt Quick are vital to the automotive IVIs, so TQtC and their partners (like KDAB, a major contributor to Qt 3D) will continue to pour lots of resources into them. In general, I believe that any effort you invest into Qt Quick and Qt 3D is quite safe for the next 5-10 years. The path forward isn't easy -- Vulkan, Metal, and DirectX 12 support are all far far behind the level of OpenGL support, so it will be years before they catch up. Nonetheless, my gut feeling tells me that dropping support for Apple is unthinkable, so TQtC will adapt to Apple's demands. And given Apple's history of viciously dropping backwards compatibility, the sense of urgency is definitely there. Some possible ways forward are: Use MoltenGL to keep the OpenGL backend usable on Apple platforms (TQtC either forms a direct partnership with Molten, or Qt customers buy MoltenGL themselves), OR Add abstraction layers to decouple Qt from OpenGL and add a Metal backend directly, OR Like #2 but provide a Vulkan backend for Qt 3D and Qt Quick, and then use MoltenVK to support Apple platforms (like how OpenGL + ANGLE is currently used to support Windows) Personally, I think #3 makes most sense because Vulkan support directly benefits Qt on all platforms, not just Apple platforms. However, I can't tell if this is a feasible business strategy for TQtC or not, or what kind of tradeoffs exist compared to a direct Metal backend. But anyway, for a more reliable source, this recent email from a TQtC employee suggests that they are currently planning the direct-Metal route: https://lists.qt-project.org/pipermail/interest/2018-December/063503.html There should be a more formal announcement sometime this year. (Disclaimer: I, too, am a mere community member who lacks a crystal ball)

I found a solution. Please see my stackoverflow post.

I saw terrain rendering in this old video, but I couldn't find the code to it: https://www.youtube.com/watch?v=bYgSxoXgSU8 Maybe you have more luck.

Never mind the problem was in my code - /I was not counting correctly the vertices :-)/ Everything works as expected .

@Cute3d, yes, it finally works. Idk what was wrong. Example project:https://github.com/Nonmant/Qt3DExtras-QText2DEntity-Example And due to god-knows-what reasons it doesn't work in debug mode. Hope, that project can help someone.

OK, seems like shadow casting on the default light might be broken. Turning it on caused the scene to crash.

Ok, solution 2 was simpler than i expected. I'll post my code below, in case someone else needs something similar. It flips a QEntity along the x-axis and the winding of the triangles. It only works with the primitive type QGeometryRenderer.Triangles. Jan void mirrorEntity(Qt3DCore::QEntity *entity) { for (QNode* node: entity->childNodes()) { if (QEntity* e = dynamic_cast<QEntity*>(node)) mirrorEntity(e); if (QGeometryRenderer* r = dynamic_cast<QGeometryRenderer*>(node)) mirrorGeometryRenderer(r); } } void mirrorGeometryRenderer(QGeometryRenderer *geometryRenderer) { QGeometry* geometry = geometryRenderer->geometry(); for (QAttribute* a : geometry->attributes()) { if (a->name() == QAttribute::defaultPositionAttributeName()) { QByteArray data = a->buffer()->data(); // flip along x-axis float *bufferContent = reinterpret_cast<float*>(data.data()); int chunkSize = a->byteStride()/sizeof(float); int positionOffset = a->byteOffset(); for (uint i=0; i<a->count()*chunkSize; i+=chunkSize) { int offsetI = i + positionOffset; bufferContent[offsetI] = -bufferContent[offsetI]; // invert x } // change winding const int size = a->byteStride(); char* tmp = new char[size]; for (int i=0; i<data.length(); i+=a->byteStride()*3) { memcpy(tmp, &data.data()[i+size], size); memcpy(&data.data()[i+size], &data.data()[i+size*2], size); memcpy(&data.data()[i+size*2], tmp, size); } delete[] tmp; a->buffer()->setData(data); break; } } }

I've got the same issue with Qt3DExtras::QText2DEntity. But I found the workaround. Look like there is a bug? in implementation of Qt3DExtras::QText2DEntity that prevents internal initialization (QText2DEntityPrivate::setScene is never called). This happens when scene is already visible and Qt3DExtras::QText2DEntity is created with parent entity passed to constructor. The workaround I used to make it work is to pass nullptr (which is default) as parent in constructor and call ->setParent(parentEntity) in separate call. m_text2dLabel = new Qt3DExtras::QText2DEntity(); m_text2dLabel->setParent(parentNode);

There has been a question here already: https://forum.qt.io/topic/85533/qt3d-qml-geometry-bounding-box. But I couldn't really find any code in the github repository. What I'd suggest is you get the QGeometryRenderer from the text entity. The geometry renderer has a QGeomemtry attribute, which in turn has a set of QAttributes. There must be an attribute with a name like "vertexPosition". You need to get the QBuffer from the attribute and retrieve the vertices starting at position byteOffset (that is an attribute of QAttribute). The next vertexSize (also an attribute of QAttribute) entries should belong to the vertex position. Then continute at position byteOffset + i * byteStride, while i being the index you're reading right now. Do this for count (also an attribute) times. This way you can get all vertex positions and calculate a bounding box based on them.

@Lukadriel Hi, One year passed and still waiting for the proper example. Did they give any tips?

@Wieland I was afraid of that. Seems like a glaring omission. Oh well, I needed to download the sources anyway to peek at Scene3D. Thanks.

@jimfar Hi, I have the same problem in my Qt application... Any solutions?

Hi! There is an example with 4 cameras, see Multi Viewport QML Example. The example uses the QML API but porting it to C++ shouldn't be too hard.

Hello, how to solve this problem? Could you please tell me the answer, thanks in advance!

I have reported this issue on the Qt bug tracker. If someone is interested: https://bugreports.qt.io/browse/QTBUG-60738