
Monocular depth unprojection (NOT multi-view triangulation). Bilinearly samples a MetricDepthMap at each [Point<Double>] and inverse-pinhole back-projects through a CameraIntrinsics into camera-frame 3D.
Unproject 2D points into camera-frame 3D and print the results for inspection. The three typed inputs (a 2D point list, a metric depth map, and camera intrinsics) feed this component, whose 3D point list is logged. Connecting a plain [Point3d<Double>] sink selects the strict arm, which asserts every point has valid depth.
Publish the recovered 3D coordinates over HTTP for a downstream consumer instead of logging them. Connecting a [Maybe<Point3d<Double>>] sink selects the optional arm, which keeps one entry per input point and so stays index-aligned with the input even when some points lack valid depth.
MetricDepthMap rather than intersecting rays from several cameras.[Point3d<Double>] asserts every input point has valid depth, while a [Maybe<Point3d<Double>>] emits one entry per input point (Some when liftable, Nothing when vetoed) so positional alignment with the input is always preserved.CameraIntrinsics message, not from configuration; intrinsics whose principal point does not match the camera bias the result without any error.Automated release