Model training
Synthetic training data, built to your spec.
Tell us what your model needs to see. We build the scenes, render them, and deliver every image with pixel labels, depth and 3D positions.
Try it: flip through a sample dataset.
Pick one of twelve angles and switch between its picture, labels, depth and boxes.
Entrance · 147 labelled objects · 15 classesCamera views
- wall
- floor
- ceiling
- window
- window casing
- door
- door casing
- fireplace
- stairs
- wall panel
- frieze
- baseboard
- chair rail
- crown molding
- ceiling border
- ceiling rose
- trim
- facade
- roof
- ground
ExploreChange the light, and watch the labels stay put.
Tell us the place. We build it.
We generate worlds, kitchens, houses and libraries, all in Dreamlands.
GenerateA factory floor or a city street, indoors or out, real or imagined: we generate the place your model needs to see. The rest of this page is one example, a country house.
Get as many angles as you need.
Get twelve cameras through the country house, or thousands.
ShootEvery camera goes inside a room automatically, at eye height and clear of the walls.
Get every pixel labelled.
See the same angles, coloured by what each part is.
LabelEvery wall, window, panel and trim takes its label from the model it was built with, so nothing is labelled by hand.
Get every object’s position in 3D.
Get a depth for every pixel, and turn any picture into a 3D point cloud.
PictureDepth, in metresLocateEvery picture ships with its camera, and every object with its exact position, rotation and size in the scene.
Vary the light and weather, keep the labels.
See one angle in morning, noon, overcast and evening light.
RelightThe picture changes. The labels don’t.
Vary every finish.
We retexture every object and match whole sets to one finish.
RestyleFabrics, woods and colours change across a scene, so your model learns the object, not its paint. The label stays the same.
In short.
- You get
- A dataset built to your spec, in your format
- A label and a depth for every pixel
- Every object’s position, rotation and size in 3D
- Works with
- Unreal Engine
- Unity
- Blender
- glTF / FBX / USD
- Your own training code