Research map / Detection and segmentation / 3D detection and driving
DeGuNet: Depth-Guided Ultra-Compact Backbones for Efficient LiDAR-Camera 3D Detection
haifa zhang, Yijing Wang, Peixi Peng, Zhiqiang Zuo
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 2D-pretrained backbone with DeGuNet | BEVFusion'23 · nuScenes val | mAP: 68.50 -> 69.40 | Improved |
| replace backbone with DeGuNet | EA-LSS · nuScenes val | mAP: 70.9 -> 71.1 | Improved |
| replace backbone with DeGuNet | GraphBEV · nuScenes val | NDS: 72.9 -> 72.9 | No clear effect |
| replace backbone with DeGuNet | IS-Fusion · nuScenes val | mAP: 71.0 -> 71.3 | Improved |
| replace backbone with DeGuNet | BEVFusion · nuScenes val (efficiency, A100) | GPU memory: 20.46GB -> 6.86GB | Improved |
| replace backbone with DeGuNet | BEVFusion · nuScenes val (efficiency, A100) | FPS: 4.4 -> 5.1 | Improved |
| pretrain generic lightweight backbones (ResNet, MobileViT) on depth completion instead of using sparsity-aware DeGuNet (ablation) | lightweight backbones · Depth pretraining on KITTI, downstream nuScenes 3D det | mAP: 62.11 / 62.09 vs 69.40 (DeGuNet) | Hurt |
| scale up to Swin-T for depth pretraining instead of sparsity-aware design (ablation) | depth-pretrained backbone · Depth pretraining on KITTI, downstream nuScenes 3D det | mAP: 64.01 vs 69.40 | Hurt |
| use DeGuNet backbone instead of auxiliary BEVDepth supervision (ablation) | BEVFusion · nuScenes val (auxiliary depth comparison) | mAP: 64.8 -> 70.3 | Improved |
| add MPIR blocks (ablation) | LiDAR baseline (BEVFusion'23) · nuScenes val (BEVFusion'23 ablation) | mAP: 64.66 -> 66.14 | Improved |
| add progressive Guide modules (ablation) | MPIR variant · nuScenes val (BEVFusion'23 ablation) | mAP: 66.14 -> 67.24 | Improved |
| add MMViT mask-aware attention (ablation) | MPIR+Guide variant · nuScenes val (BEVFusion'23 ablation) | mAP: 67.24 -> 68.38 | 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.
- IS-Fusion — IS-Fusion: Instance-Scene Collaborative Fusion for Multimodal 3D Object Detection (CVPR 2024) · better on 2 of 2 shared results
- GraphBEV — GraphBEV: Towards Robust BEV Feature Alignment for Multi-Modal 3D Object Detection (ECCV 2024) · better on 1 of 1 shared result
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
- RayD3D: Distilling Depth Knowledge Along the Ray for Robust Multi-View 3D Object Detection (AAAI 2026 · similarity 0.87)
- Advancing 3D Object Detection With Depth-Aware Spatial Knowledge Distillation (TPAMI 2025 · similarity 0.87)
- Multi-View 3D Object Detection Network for Autonomous Driving (CVPR 2017 · similarity 0.87)
- SimDistill: Simulated Multi-Modal Distillation for BEV 3D Object Detection (AAAI 2024 · similarity 0.86)
- PromptDet: A Lightweight 3D Object Detection Framework with LiDAR Prompts (AAAI 2025 · similarity 0.86)
- SwiftPillars: High-Efficiency Pillar Encoder for Lidar-Based 3D Detection (AAAI 2024 · 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.
