AngelSeen’s Public Lab for Spatial Vision

Field notes

AngelSeen separates camera ingestion from spatial understanding, combining reconstruction, pose and device registration with detection, tracking, simulation, and vision-language models. Its contract with application developers is structured output rather than raw video: scenes, cameras, objects, trajectories, semantics, and events. The first experiment models a store with three registered cameras and a semantic map covering locations such as the entrance, main aisle, displays, and checkout. The page reports 12 current objects and 47 zone events in its rendered readout.