Skip to main content

Three.js Renderer

In the WebGL custom renderer guide we rendered a cube using the WebGL API directly. This worked well, but we had to write a lot of code to create and bind buffers, compiling shaders etc. With a higher level 3D library like Three.js we can achieve the same thing but a bit more effiently.

You will learn
  • How to use Three.js to render 3D objects directly into the MapillaryJS rendering context
  • How to implement the ICustomRenderer interface by leveraging the Three.js functionality
  • How to add your renderer to the Viewer

Creating a 3D Object#

We will use the Three.js BoxGeometry to define the shape of our cube. We use an array of materials to give each face of the cube Mesh a different color.

function makeCubeMesh() {  const geometry = new BoxGeometry(2, 2, 2);  const materials = [    new MeshBasicMaterial({      color: 0xffff00,    }),    new MeshBasicMaterial({      color: 0xff00ff,    }),    new MeshBasicMaterial({      color: 0x00ff00,    }),    new MeshBasicMaterial({      color: 0x0000ff,    }),    new MeshBasicMaterial({      color: 0xffffff,    }),    new MeshBasicMaterial({      color: 0xff0000,    }),  ];  return new Mesh(geometry, materials);}

Geo Coordinates#

Entities rendered in MapillaryJS have geodetic coordiantes (or have a position relative to a geodetic reference coordinate) so we set the cube's geo position. The mesh position is initialized to the origin, we will soon change that.

const cube = {  geoPosition: {    alt: 1,    lat: -25.28268614514251,    lng: -57.630922858385,  },  mesh: makeCubeMesh(),};

That's everything we need to start implementing our custom renderer.

Creating the Custom Renderer#

To create our custom renderer, we will implement the ICustomRenderer interface.

Let us go through it member by member.


We can use the constructor to assign readonly properties of our renderer. Every custom renderer needs to have a unique ID and specify its render pass (currently the only supported render pass is Opaque). Our cube will also be readonly so we assign it in the constructor as well.

class ThreeCubeRenderer {  constructor(cube) { = 'three-cube-renderer';    this.renderPass = RenderPass.Opaque;    this.cube = cube;  }  // ...}


ICustomRenderer.onAdd is called when the renderer has been added to the Viewer with the Viewer.addCustomRenderer method. This gives your renderer a chance to initialize resources and register event listeners. It is also a chance to calculate the local topocentric positions for scene objects using the provided reference.

To calculate the topocentric position of our cube, we will make use of the geodeticToEnu helper function in MapillaryJS and make a position array.

function geoToPosition(geoPosition, reference) {  const enuPosition = geodeticToEnu(    geoPosition.lng,,    geoPosition.alt,    reference.lng,,    reference.alt,  );  return enuPosition;}

With our helper function, we can calculate and set the local topocentric position of our cube by using the reference parameter for the conversion. After that, we will make use of the Three.js Camera, Scene and WebGLRenderer classes to prepare for the animation frames where we will render the cube.

We first get the Viewer canvas and use the context parameter to setup our renderer. We do not want to clear the canvas before rendering our cube (that would erase the street imagery visualization) so we deactivate the autoClear functionality.

After that we create the camera we will use for rendering. We will manually update the viewMatrix of the camera so we deactivate the matrixAutoUpdate functionality.

At last, we create our scene and add the cube mesh to it.

class ThreeCubeRenderer {  // ...  onAdd(viewer, reference, context) {    const position = geoToPosition(this.cube.geoPosition, reference);    this.cube.mesh.position.fromArray(position);
    const canvas = viewer.getCanvas();    this.renderer = new WebGLRenderer({      canvas,      context,    });    this.renderer.autoClear = false; = new Camera(); = false;
    this.scene = new Scene();    this.scene.add(this.cube.mesh);  }}


While we will only operate in a small area around the cube with our renderer, MapillaryJS operates on global earth scale. For different reasons, e.g. to ensure numeric stability by keeping topocentric coordinates sufficiently small, MapillaryJS will sometimes update its internal reference geo coordinate used to convert coordinates from geodetic to local topocentric reference. Whenever it updates the reference, it will notify our renderer by calling the ICustomRenderer.onReference method so that we can act accordingly and recalculate our cube position. This does not mean the the cube moves relative to the street imagery. Instead, the earth sphere that MapillaryJS operates on has been rotated and we need to adjust everything we want to render accordingly.

class ThreeCubeRenderer {  //...  onReference(viewer, reference) {    const position = geoToPosition(this.cube.geoPosition, reference);    this.cube.mesh.position.fromArray(position);  }}


ICustomRenderer.onRemove is called when the renderer has been removed from the Viewer with the Viewer.removeCustomRenderer method. This gives us a chance to clean up our Three.js resources (and potential event listeners etc). Everything that we created in the onAdd method should be disposed and cleaned up now.

class ThreeCubeRenderer {  //...  onRemove(_viewer, _context) {    this.cube.mesh.geometry.dispose();    this.cube.mesh.material.forEach((m) => m.dispose());    this.renderer.dispose();  }}


ICustomRenderer.render is called during every animation frame that is run. It allows our renderer to draw into the WebGL context.


When the Viewer is halted, i.e. when no motion such as translation or rotation is performed, the animation frames are not run and therefore the render method will not be called. See the Animation example for guidance into how to force all animation frames to be run and the render method to be called on every frame.

The render method provides the viewMatrix and projectionMatrix as parameters. The viewMatrix inverse is equivalent to the Three.js camera matrix so we set the camera matrix from our array and invert it using the Three.js matrix math. After setting the camera matrix it is important to make a call to updateMatrixWorld to ensure that all the internal camera matrices are synchronized. We also set the camera projection matrix from our projectionMatrix array.

Finally, we reset the renderer state to ensure that previous buffers etc. are cleared before our call to render the scene with our camera.

class ThreeCubeRenderer {  //...  render(context, viewMatrix, projectionMatrix) {    const {camera, scene, renderer} = this;    camera.matrix.fromArray(viewMatrix).invert();    camera.updateMatrixWorld(true);    camera.projectionMatrix.fromArray(projectionMatrix);
    renderer.resetState();    renderer.render(scene, camera);  }}

Additional Functionality#

In our custom renderer we only have a single fixed cube that is always visible. Maybe you want to add and remove objects dynamically, change object positions, or change object appearance during the lifespan of the renderer and application. To do that, you can add additional methods and functionality to your renderer class directly or in helpers.


As of this writing, MapillaryJS will always render the street imagery layer as a background. Occlusion between custom rendered objects and the street imagery will never occur, custom rendered objects will always be rendered on top of the street imagery. You can eperiment with transparency to assert object placement.

Adding the Renderer#

Now that we have implemented our custom cube renderer, we just need to add it to the Viewer through the Viewer.addCustomRenderer method.


Try changing the cube's geometry and face colors.

Live Editor
SyntaxError: Unexpected token (1:8)
1 : return ()


  • To add your Three.js 3D objects to MapillaryJS, implement the ICustomRenderer interface
  • Make sure your objects have a geo position (or a position relative to a geo reference)
  • Add your objects to a Three.js Scene in your renderer
  • Use the Three.js Camera, and WebGLRenderer classes to render directly into the MapillaryJS rendering context
  • Add your custom renderer to the Viewer

You can view the complete code in the Three.js Renderer example.

For more examples of custom renderers using Three.js, you can take a look at the OpenSfM axes and earth grid renderers.