Research map / Detection and segmentation / 3D detection and driving
Towards Accurate and Efficient 3D Object Detection for Autonomous Driving: A Mixture of Experts Computing System on Edge
Linshen Liu, Boyan Su, Junyue Jiang, Guanlin Wu, Cong Guo, Ceyu Xu, Hao Frank Yang
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 |
|---|---|---|---|
| add the VEE expert to the MoE system (ablation) | EMC2 · KITTI val Car | 3D AP: 86.40 -> 90.50 | Improved |
| add the VEE expert to the MoE system (ablation) | EMC2 · KITTI val latency on Jetson | latency (ms): 512.7 -> 372.5 | Improved |
| use APE (with image features) instead of VEE in hard scenes (ablation) | EMC2 · KITTI val Car | 3D AP: 87.69 -> 90.50 | Improved |
| add multiscale pooling to hierarchical training (ablation) | EMC2 · KITTI val latency on Jetson | latency (ms): 1320 -> 526 | Improved |
| add hierarchical training (ablation) | EMC2 · KITTI val Car | 3D AP: 88.68 -> 90.50 | Improved |
| use LPE instead of heavier experts in close-range scenes (ablation) | EMC2 · KITTI val LPE scenes | latency (ms): 526 -> 219 | Improved |
| route scenes to multimodal MoE experts versus using LiDAR-only or multimodal baselines | EMC2 · KITTI Cyclist Hard | 3D AP: 72.42 -> 77.62 (vs LoGoNet) | Improved |
| EMC2 versus 3ONet on Car Easy and TED on Cyclist Easy | EMC2 · KITTI Car and Cyclist Easy | 3D AP: 94.24 vs 94.00 (Car Easy) | Hurt |
Closest papers by meaning
- Toward Optimal Mixture of Experts System for 3D Object Detection: A Game of Accuracy, Efficiency and Adaptivity (TPAMI 2025 · similarity 0.93)
- DAIR-V2X: A Large-Scale Dataset for Vehicle-Infrastructure Cooperative 3D Object Detection (CVPR 2022 · similarity 0.87)
- Ev-3DOD: Pushing the Temporal Boundaries of 3D Object Detection with Event Cameras (CVPR 2025 · similarity 0.86)
- DI-V2X: Learning Domain-Invariant Representation for Vehicle-Infrastructure Collaborative 3D Object Detection (AAAI 2024 · similarity 0.85)
- SwiftPillars: High-Efficiency Pillar Encoder for Lidar-Based 3D Detection (AAAI 2024 · similarity 0.85)
- CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes (CVPR 2025 · similarity 0.85)
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.
