OccAny: Generalized Unconstrained Urban 3D Occupancy
Researchers have introduced OccAny, a novel framework for predicting and completing 3D urban occupancy from uncalibrated, out-of-domain images. This system is designed to overcome the limitations of existing methods that rely on in-domain annotations and precise sensor priors. OccAny utilizes a Segmentation Forcing technique to enhance occupancy quality and enable mask-level predictions, alongside a Novel View Rendering pipeline for geometry completion through test-time view augmentation. Experiments show OccAny surpasses visual geometry baselines and rivals in-domain methods on urban occupancy prediction tasks. AI