Save custom geometry to obj format

Is there currently any way of saving custom geometry to file? For example to obj format? So far I can think only of using Assimp.

Try WriteDrawablesToOBJ() function in Drawable.h.

1 Like

I am getting empty object if I call it from the Drawable itself. What would be the correct way? This is my code called from inside a custom drawable.

PODVector<Drawable*> drawables;
String path = "/test.obj";
SharedPtr<File> file = SharedPtr<File>(new File(context_, path, FILE_WRITE));

Recommend debugging inside the function. It does IsEnabledEffective() check right in the start, which could result to no output if the drawable / its node is not enabled.

It is enabled I am drawing RibbonTrail and I have the above code assigned to a hotkey, trying to capture the resulted geometry for debugging. I made my own copy of RibbonTrail and I am calling the code from inside passing this pointer as a drawable.

The drawable write will only work on geometries whose buffers have CPU-side shadow data enabled, like ordinary models. RibbonTrail does constant dynamic regeneration, so for it it would be just a performance & memory hog.

I understand it would be slow but I want to capture just a static state at one frame. How can I modify and copy the data in order to save it?

You could maybe go through the Drawable’s Geometries, and the Vertex / IndexBuffers it uses, and call SetShadowed(true) on all of them. The next update would then copy the vertex/index data also to main memory for capture. I don’t promise this will work.

It works partially, but the data is a bit scrambled every corner of the triangle is collapsed to a single point. When I import the obj I am getting a line where multiple points lie on top of each other.

Part of the work in getting e.g. ribbontrails and billboards to render correctly is in the vertex shader. Therefore just a dump of the vertex data, interpreting positions as just positions and multiplying them with the object’s model matrix, isn’t necessarily going to work 100%.

It would be the same with skeletal animation - Urho doesn’t contain CPU skinning code at all and relies on the vertex shader, so you couldn’t dump an AnimatedModel in its actual pose.

Thanks cadaver. Can you point me to the GLSL code doing this? I can’t find any relevant code, It is just using the Unlit shader.

The actual vertex transform happens sort of “behind the scenes”, as the iModelMatrix is actually a macro and does magic depending on geometry type. Every shader including Unlit typically does mat4 modelMatrix = iModelMatrix; What actually happens, check Transform.glsl, from about line 147 onward.

I see is it using the TANGENT in the shader for the transformation? Is there any benefit of doing this in the shader?

As I understand it, Billboard (for instance) packs a rotation matrix in the vertex buffer, which is possible because a bunch of added space has been added via setting the TEXCOORD flags.

            float rotationMatrix[2][2];
            SinCos(billboard.rotation_, rotationMatrix[0][1], rotationMatrix[0][0]);
            rotationMatrix[1][0] = -rotationMatrix[0][1];
            rotationMatrix[1][1] = rotationMatrix[0][0];

            dest[0] = billboard.position_.x_;
            dest[1] = billboard.position_.y_;
            dest[2] = billboard.position_.z_;
            ((unsigned&)dest[3]) = color;
            dest[4] = billboard.uv_.min_.x_;
            dest[5] = billboard.uv_.min_.y_;
            dest[6] = -size.x_ * rotationMatrix[0][0] + size.y_ * rotationMatrix[0][1];
            dest[7] = -size.x_ * rotationMatrix[1][0] + size.y_ * rotationMatrix[1][1];

Then, GraphicsDefs.h exposes some enums (FC_DIRECTION,etc) that are then handled in Transform.glsl.

IMO, it is pretty clever to use the vertex structures in this way, but it makes for a pretty opaque code base. I wonder if there is a better/clearer way.

I think for Billboard it makes sense but for RibbonTrail I am confused as there is no instancing involved.