Research map / Detection and segmentation / 3D detection and driving
Height-Fidelity Dense Global Fusion for Multi-modal 3D Object Detection
Hanshi Wang, Jin Gao, Weiming Hu, Zhipeng Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace the window transformer with vanilla Mamba (ablation) | UniTR · nuScenes (50% subset) | mAP: 70.6 -> 65.8 | Hurt |
| add the hybrid block with height-fidelity encoding to Mamba fusion (ablation) | vanilla Mamba fusion · nuScenes (50% subset) | mAP: 65.8 -> 71.9 | Improved |
| add our design to RetNet instead of Mamba (ablation) | RetNet fusion · nuScenes (50% subset) | mAP: 64.6 -> 70.7 | Improved |
| add Hybrid Mamba Block in raw 3D space without height-fidelity encoding (ablation) | Baseline-LC · nuScenes (50% subset) | mAP: 67.7 -> 66.9 | Hurt |
| add height-fidelity LiDAR encoding to raw-space HMB (ablation) | Baseline-LC + HMB-R · nuScenes (50% subset) | mAP: 66.9 -> 70.3 | Improved |
| add global Mamba to local-only Hybrid Mamba Block (ablation) | Hybrid Mamba Block · nuScenes (50% subset) | mAP: 70.5 -> 71.9 | Improved |
| add local Mamba to global-only Hybrid Mamba Block (ablation) | Hybrid Mamba Block · nuScenes (50% subset) | mAP: 71.5 -> 71.9 | Improved |
| use Hilbert curve instead of Z-order curve for sorting (ablation) | Global Mamba · nuScenes (50% subset) | mAP: 71.5 -> 71.9 | Improved |
| use Hilbert curve instead of coordinate order (ablation) | Global Mamba · nuScenes (50% subset) | mAP: 71.2 -> 71.9 | Improved |
| add xy-direction serialization in local Mamba (ablation) | Hybrid Mamba Block · nuScenes (50% subset) | mAP: 69.5 -> 70.9 | Improved |
| add bidirectional scanning in global Mamba (ablation) | Hybrid Mamba Block · nuScenes (50% subset) | mAP: 70.9 -> 71.9 | Improved |
| add height-aware encoding in the LiDAR backbone (ablation) | Hybrid Mamba fusion · nuScenes (50% subset) | mAP: improves the mAP by 0.9 and NDS by 0.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- UniTR — UniTR: A Unified and Efficient Multi-Modal Transformer for Bird's-Eye-View Representation (ICCV 2023) · better on 4 of 4 shared results
- IS-FUSION — IS-Fusion: Instance-Scene Collaborative Fusion for Multimodal 3D Object Detection (CVPR 2024) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Fully Sparse Fusion for 3D Object Detection (TPAMI 2024 · similarity 0.88)
- CCF: Complementary Collaborative Fusion for Domain Generalized Multi-Modal 3D Object Detection (CVPR 2026 · similarity 0.88)
- Look Before You Fuse: 2D-Guided Cross-Modal Alignment for Robust 3D Detection (CVPR 2026 · similarity 0.88)
- SimDistill: Simulated Multi-Modal Distillation for BEV 3D Object Detection (AAAI 2024 · similarity 0.88)
- UniMamba: Unified Spatial-Channel Representation Learning with Group-Efficient Mamba for LiDAR-based 3D Object Detection (CVPR 2025 · similarity 0.88)
- Spatial-Mamba: Effective Visual State Space Models via Structure-Aware State Fusion (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
