Towards 3D Objectness Learning in an Open World by Taichi Liu, Zhenyu Wang, Ruofeng Liu, Guang Wang, Desheng Zhang instant download

Towards 3D Objectness Learning in an Open World by Taichi Liu, Zhenyu Wang, Ruofeng Liu, Guang Wang, Desheng Zhang instant download

arXiv:2510.17686v1 [cs.CV] 20 Oct 2025AbstractRecent advancements in 3D object detection and novel category detection havemade significant progress, yet research on learning generalized 3D objectnessremains insufficient. In this paper, we delve into learning open-world 3D objectness,which focuses on detecting all objects in a 3D scene, including novel objects unseenduring training. Traditional closed-set 3D detectors struggle to generalize to openworld scenarios, while directly incorporating 3D open-vocabulary models for openworld ability struggles with vocabulary expansion and semantic overlap. To achievegeneralized 3D object discovery, we propose OP3Det, a class-agnostic OpenWorld Prompt-free 3D Detector to detect any objects within 3D scenes withoutrelying on hand-crafted text prompts. We introduce the strong generalizationand zero-shot capabilities of 2D foundation models, utilizing both 2D semanticpriors and 3D geometric priors for class-agnostic proposals to broaden 3D objectdiscovery. Then, by integrating complementary information from point cloudand RGB image in the cross-modal mixture of experts, OP3Det dynamicallyroutes uni-modal and multi-modal features to learn generalized 3D objectness.Extensive experiments demonstrate the extraordinary performance of OP3Det,which significantly surpasses existing open-world 3D detectors by up to 16.0% inAR and achieves a 13.5% improvement compared to closed-world 3D detectors.1 IntroductionIn 3D perception systems, especially in real-world environments such as autonomous driving androbotics, object categories of interest may change dynamically. This has led to increasing attention onchallenging tasks like out-of-distribution 3D detection [1, 2], open-world 3D detection [3, 4] and openvocabulary 3D detection [5, 6, 7, 8, 9, 10], improving generalization beyond closed-set assumptions.A core challenge across these tasks is the ability to localize all objects, which lies in understandinghow objects are structured in 3D