Segmentation explained: object masks and the visible road

A box and a mask answer different questions

A detector places a rectangle around a possible object. An instance mask estimates which pixels belong to that particular object, helping separate its visible shape from the background. A mask is still a prediction: thin parts, reflections and partially hidden surfaces can be inaccurate. It is not a complete 3D model of the object.

Choose nawnaw seg pro

Open the AI model picker and select nawnaw seg pro with Pro access. The app routes this model to its dedicated Segmentation page, where road-surface and object-mask outputs can be displayed. If you want the Detection, Maps, Virtual and Parking pages again, select nawnaw or nawnaw pro. A model choice changes the available output, not just the colour theme.

What the road surface means

The road output estimates a visible road region in the camera image. It can make the distinction between an object and the surface around it easier to read. It does not certify that the entire region is safe to drive on. Slopes, shadows, glare and occlusion can make the boundary incomplete or ambiguous.

What it does not identify

A road mask does not establish lane count, a pedestrian crossing, a stop line or a building facade. The current model is not a dedicated building segmentation model, and it does not provide reliable vehicle nose direction. Buildings in 3D Maps are map data, not facades recovered from these masks. Keep these sources separate when interpreting the app.

Choose detail within the device budget

Masks add work beyond displaying detection boxes. Real responsiveness depends on the iPhone, temperature, recording and other active features. Compare models while stationary and use the app’s power controls when needed. If you only need boxes, a detector is the more direct tool. No fixed scan rate should be read as a guarantee of segmentation performance.