To build realistic MR apps, both Scene and Depth API capabilities should be used to cover a broader set of use cases than either capability can address in isolation.

The recommended flow is to:

If space setup is not initiated, the Depth API cannot take advantage of the scene model when computing the depth maps. In this case, the depth maps returned from the Depth API will be computed only from the sensed depth. This will make the depth estimates worse and less stable, especially on planar surfaces such as walls, floors and tables. When using the Depth API for occlusion, this will increase flickering when virtual objects are close to these real-world surfaces.

It is technically possible to use Depth API independently of Scene. Note that when done so, it will only expose depth maps based on sensed depth alone. Depending on the use case for your app, you may choose to only use the Depth API.