Research map / Point clouds / Segmentation and detection
MV2DFusion: Leveraging Modality-Specific Object Semantics for Multi-Modal 3D Detection
Zitian Wang, Zehao Huang, Yulu Gao, Naiyan Wang, Si Liu
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 |
|---|---|---|---|
| raise image resolution from 704x256 to 1600x640 (ablation) | MV2DFusion · nuScenes val | NDS: 0.740 -> 0.745 | Improved |
| use ConvNext-L backbone instead of ResNet-50 (ablation) | MV2DFusion · nuScenes val | NDS: 0.747 -> 0.755 | Improved |
| swap 2D detector (YOLOX, Faster R-CNN, Cascade R-CNN) (ablation) | MV2DFusion · nuScenes val | NDS: 0.745 -> 0.747 | No clear effect |
| remove camera input at test time | MV2DFusion trained with both · nuScenes val | NDS: 0.747 -> 0.469 | Hurt |
| add mixed-modality training (ablation) | MV2DFusion · nuScenes val, camera only | NDS: 0.469 -> 0.523 | Improved |
| add mixed-modality training (ablation) | MV2DFusion · nuScenes val, camera and LiDAR | NDS: 0.747 -> 0.748 | Improved |
| use distribution instead of point image queries (ablation) | MV2DFusion · nuScenes val | NDS: 0.743 -> 0.747 | Improved |
| use distribution instead of point image queries (ablation) | MV2DFusion · AV2 val | CDS: 0.373 -> 0.395 | Improved |
| add image queries to LiDAR-only queries (ablation) | MV2DFusion · nuScenes val | NDS: 0.737 -> 0.747 | Improved |
| add LiDAR queries to image-only queries (ablation) | MV2DFusion · nuScenes val | NDS: 0.733 -> 0.747 | Improved |
| add cross-attention to decoder (ablation) | MV2DFusion · nuScenes val | mAP: 0.715 -> 0.728 | Improved |
| increase decoder layers from 3 to 6 (ablation) | MV2DFusion · nuScenes val | NDS: 0.746 -> 0.747 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BEVFusion — BEVFusion: A Simple and Robust LiDAR-Camera Fusion Framework (NeurIPS 2022) · better on 2 of 2 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
- CCF: Complementary Collaborative Fusion for Domain Generalized Multi-Modal 3D Object Detection (CVPR 2026 · similarity 0.91)
- MGAF: LiDAR-Camera 3D Object Detection With Multiple Guidance and Adaptive Fusion (TPAMI 2025 · similarity 0.90)
- SimDistill: Simulated Multi-Modal Distillation for BEV 3D Object Detection (AAAI 2024 · similarity 0.89)
- Adaptive-Smooth LiDAR-Camera Knowledge Distillation with Heterogeneous Fusion for Multi-View 3D Object Detection (AAAI 2026 · similarity 0.89)
- Homogeneous Multi-modal Feature Fusion and Interaction for 3D Object Detection (ECCV 2022 · similarity 0.89)
- MSMDFusion: Fusing LiDAR and Camera at Multiple Scales With Multi-Depth Seeds for 3D Object Detection (CVPR 2023 · similarity 0.88)
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.
