OccAny: Generalized Unconstrained Urban 3D Occupancy

A new computer vision model called OccAny can predict three-dimensional occupancy in urban environments without requiring the calibrated sensor setups that constrain existing systems, according to a paper posted on the arXiv preprint server [1]. The work, submitted on 24 March 2026 by researcher Anh-Quan Cao and collaborators, describes what the authors call "the first unconstrained urban 3D occupancy model" [1][2]. Unlike current methods that depend on in-domain annotations and precise sensor-rig priors, OccAny operates on out-of-domain, uncalibrated scenes [2]. The model accepts three input configurations — sequential, monocular, or surround-view images — and outputs metric occupancy predictions coupled with segmentation features [1][2]. The paper introduces three technical contributions: a generalized 3D occupancy framework, a technique named Segmentation Forcing that improves occupancy quality while enabling mask-level prediction, and a Novel View Rendering pipeline that infers novel-view geometry to support test-time view augmentation for geometry completion [2]. In experiments across two established urban occupancy prediction datasets, OccAny outperformed all visual geometry baselines and remained competitive with in-domain self-supervised methods [1][2]. The submission file weighed 47,018 KB and was last revised on 11 June 2026 [1]. The paper appears on arXiv, an open-access repository of electronic preprints that, as of November 2024, receives about 24,000 articles per month and hosts more than two million papers across disciplines including computer science, physics, and mathematics [6]. arXiv preprints are moderated but not peer-reviewed [6]. The article’s abstract page features integrations from arXivLabs, a framework launched in 2020 that allows community collaborators to develop experimental tools on the platform [5]. These include the Bibliographic Explorer for citation-tree navigation, the CORE Recommender for discovering related open-access papers, and Connected Papers for visualizing citation networks [4][5]. arXivLabs partners must adhere to values of openness, community, excellence, and user data privacy, and are granted only minimal, anonymized user data [5]. The framework is currently on hiatus for new proposals while the arXiv development team focuses on migrating systems to the cloud [3].