Search papers, labs, and topics across Lattice.
This paper introduces STAR, a Spatial-Topology Aware Routing Framework designed to enhance 3D scene understanding by addressing the challenges posed by topological discrepancies across sensor modalities. By integrating a multi-attribute self-supervised pre-training branch and a domain-aware expert branch that employs Domain-Spatial-Guided Routing and Entropy-controlled Dynamic Allocation, STAR effectively captures local topological variations while ensuring adaptive expert allocation. Experimental results show that STAR significantly outperforms strong baselines, achieving 80.1% mIoU on the ScanNet validation set and 77.2% mIoU on S3DIS, highlighting its robustness across diverse 3D scenes.
STAR achieves state-of-the-art performance in 3D scene understanding by effectively balancing semantic consistency with geometric heterogeneity.
Constructing a unified 3D scene understanding model has long been hindered by the topological discrepancies across sensor modalities. While applying the Mixture-of-Experts (MoE) architecture is a flexible approach for multi-domain 3D understanding, we observe that conventional feature-only MoE routers may underrepresent local sampling topology under semantic supervision, making expert allocation difficult when semantic consistency coexists with geometric heterogeneity. To overcome this challenge, we propose STAR (Spatial-Topology Aware Routing Framework). Specifically, we introduce a multi-attribute self-supervised pre-training branch, covering topological and textural variations, to anchor cross-domain structural priors. Building upon this, we design a domain-aware expert branch with two mechanisms: Domain-Spatial-Guided Routing (DSR), which captures local topological variations from spatial context, and Entropy-controlled Dynamic Allocation (EDA), which adjusts the number of activated experts according to routing uncertainty. Together, these branches combine stable cross-domain representation learning with adaptive expert allocation. Extensive experiments across various tasks, encompassing both indoor and outdoor scenes, demonstrate the effectiveness of STAR. It achieves 80.1% mIoU on the ScanNet validation set and 77.2% mIoU on S3DIS, consistently improving over strong baselines. Code is available at our project page (https://xmw666.github.io/STAR/).
