What a camera produces: format and geometry.
There are two of these in play for any running camera, and the difference matters.
The declared one comes from BB.Camera.stream_info/1, which runs during
discovery against robots that aren't started. It can only describe what the
DSL asked for.
The live one is whatever the driver handed to BB.Camera.Hub once it had
opened the device and seen a real frame. A camera asked for 1280x720 that
negotiated 640x480 reports the latter, as soon as it knows it, via
BB.Camera.Hub.update_stream_info/2.
Consumers are told about a change before the first frame it describes, so nothing ever attributes a frame to the wrong geometry.
Summary
Types
@type t() :: %BB.Camera.StreamInfo{ format: BB.Camera.Format.t(), frame_rate: float() | nil, height: pos_integer(), metadata: map(), width: pos_integer() }