mlsanityData to 14 Oct 2026

STAR: A Spatial-Topology Aware Routing Framework for Generalizable 3D Scene Understanding

arXiv 2608.11699 ↗First posted 12 Aug 2026cs.CVcited by 0

On the map: Mixture-of-experts → Vision mixture-of-experts

Abstract (arXiv)

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/).

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Vision mixture-of-experts, one of 12 directions in Mixture-of-experts. Builds on 52 of its 63 references that are in the map. First posted on arXiv 12 Aug 2026.

Builds on

52 of 63 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Uni3D-MoE: Scalable Multimodal 3D Scene Understanding via Mixture of Experts27 May
  2. 2025LiMoE: Mixture of LiDAR Representation Learners from Automotive Scenes7 Jan
  3. 2022Sparse Mixture-of-Experts are Domain Generalizable Learners8 Jun
  4. 2021Generalizable Person Re-identification with Relevance-aware Mixture of Experts19 May

Other references · 48

  1. 2018Representation Learning with Contrastive Predictive Codingcited by 7,777
  2. 2016PointNet: Deep Learning on Point Sets for 3D Classification and Segmentationcited by 5,248
  3. 2019nuScenes: A multimodal dataset for autonomous drivingcited by 4,523
  4. 2017PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Spacecited by 4,401
  5. 2017ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenescited by 3,117
  6. 2021Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsitycited by 3,073
  7. 2020Unsupervised Learning of Visual Features by Contrasting Cluster Assignmentscited by 2,857
  8. 2019Scalability in Perception for Autonomous Driving: Waymo Open Datasetcited by 2,451
  9. 2018PointPillars: Fast Encoders for Object Detection from Point Cloudscited by 1,715
  10. 2017Matterport3D: Learning from RGB-D Data in Indoor Environmentscited by 1,376
  11. 2017VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detectioncited by 1,323
  12. 2019KPConv: Flexible and Deformable Convolution for Point Cloudscited by 1,205
36 more
  1. 2019PV-RCNN: Point-Voxel Feature Set Abstraction for 3D Object Detectioncited by 891
  2. 2023Point Transformer V3: Simpler, Faster, Strongercited by 589
  3. 2020PointContrast: Unsupervised Pre-training for 3D Point Cloud Understandingcited by 504
  4. 2022Masked Autoencoders for Point Cloud Self-supervised Learningcited by 459
  5. 2021ARKitScenes: A Diverse Real-World Dataset For 3D Indoor Scene Understanding Using Mobile RGB-D Datacited by 454
  6. 2022PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategiescited by 436
  7. 2022Point Transformer V2: Grouped Vector Attention and Partition-based Poolingcited by 388
  8. 20203D-FRONT: 3D Furnished Rooms with layOuts and semaNTicscited by 380
  9. 2020PointGroup: Dual-Set Point Grouping for 3D Instance Segmentationcited by 321
  10. 2019Structured3D: A Large Photo-realistic Dataset for Structured 3D Modelingcited by 286
  11. 2023Uni3D: Exploring Unified 3D Representation at Scalecited by 237
  12. 2018Spreading vectors for similarity searchcited by 118
  13. 2022Habitat-Matterport 3D Semantics Datasetcited by 110
  14. 2025Sonata: Self-Supervised Learning of Reliable Point Representationscited by 110
  15. 2023Towards Large-scale 3D Representation Learning with Multi-dataset Point Prompt Trainingcited by 86
  16. 2024ViT-CoMer: Vision Transformer with Convolutional Multi-scale Feature Interaction for Dense Predictionscited by 66
  17. 2025SpatialLM: Training Large Language Models for Structured Indoor Modelingcited by 65
  18. 2022Meta-DMoE: Adapting to Domain Shift by Meta-Distillation from Mixture-of-Expertscited by 50
  19. 2025Concerto: Joint 2D-3D Self-Supervised Learning Emerges Spatial Representationscited by 45
  20. 2024RT-DETRv3: Real-time End-to-End Object Detection with Hierarchical Dense Positive Supervisioncited by 32
  21. 2024UniDet3D: Multi-dataset Indoor 3D Object Detectioncited by 20
  22. 2023MonoLSS: Learnable Sample Selection For Monocular 3D Detectioncited by 19
  23. 2023DUSA: Decoupled Unsupervised Sim2Real Adaptation for Vehicle-to-Everything Collaborative Perceptioncited by 17
  24. 2022PointInst3D: Segmenting 3D Instances by Pointscited by 16
  25. 2023Open-TransMind: A New Baseline and Benchmark for 1st Foundation Model Challenge of Intelligent Transportationcited by 14
  26. 2023MDT3D: Multi-Dataset Training for LiDAR 3D Object Detection Generalizationcited by 14
  27. 2025CADTrack: Learning Contextual Aggregation with Deformable Alignment for Robust RGBT Trackingcited by 11
  28. 2024One for All: Multi-Domain Joint Training for Point Cloud Based 3D Object Detectioncited by 8
  29. 2025LLaVA-SP: Enhancing Visual Representation with Visual Spatial Tokens for MLLMscited by 8
  30. 2025Sat2RealCity: Geometry-Aware and Appearance-Controllable 3D Urban Generation from Satellite Imagerycited by 7
  31. 2024RopeBEV: A Multi-Camera Roadside Perception Network in Bird's-Eye-Viewcited by 5
  32. 2026ArtifactWorld: Scaling 3D Gaussian Splatting Artifact Restoration via Video Generation Modelscited by 5
  33. 2026You Only Gaussian Once: Controllable 3D Gaussian Splatting for Ultra-Densely Sampled Scenescited by 4
  34. 2026PanoWorld: A Generative Spatial World Model for Consistent Whole-House Panorama Synthesiscited by 3
  35. 2026RAGTrack: Language-aware RGBT Tracking with Retrieval-Augmented Generationcited by 2
  36. 2026Pano2World: End-to-End 3D Generation via Unified Multi-View Sequencescited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025DoReMi: Bridging 3D Domains via Topology-Aware Domain-Representation Mixture of Experts
  2. 2026SpaR3D-MoE: Adaptive 3D Spatial Reasoning from Sparse Views Meets Geometry-Inductive Mixture-of-Experts
  3. 2026CitySTAR: Structured and Topology-Aware Reasoning for Open-Vocabulary Urban 3D Grounding
  4. 2026CAIRN: Cross-Room 3D Scene Understanding with Topology-Aware Large Multimodal Models
  5. 2026STAR: Rethinking MoE Routing as Structure-Aware Subspace Learning
  6. 2026SceneBench: A Hierarchical Benchmark for Vision-Language Understanding of 3D Scenes
  7. 2026Uncovering and Shaping the Latent Representation of 3D Scene Topology in Vision-Language Models
  8. 2026Generation of High-Level Concepts in 3D Scene Graphs via Autoregressive Diffusion