Research map / Detection and segmentation / 3D detection and driving
Detecting As Labeling: Rethinking LiDAR-camera Fusion in 3D Object Detection
Junjie Huang, Yun Ye, Zhujin Liang, Yi Shan, Dalong Du
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 BEVFusion-style pipeline with DAL pipeline (no image features in regression) plus auxiliary head, under simple training (ablation) | LiDAR-camera fusion · nuScenes val | NDS: 69.00 -> 69.52 vs LiDAR-only | Improved |
| train BEVFusion pipeline with simple ImageNet-only training (ablation) | BEVFusion · nuScenes val | NDS: 69.00 -> 68.71 | No clear effect |
| add large-range image resize augmentation (ablation) | DAL · nuScenes val | mAP / NDS: +3.91 mAP and +1.35 NDS | Improved |
| add velocity augmentation of static objects (ablation) | DAL · nuScenes val | mAP / NDS: +0.47 mAP and +1.07 NDS; mAVE 25.80 -> 19.31 | Improved |
| add auxiliary classification head to BEVFusion-like pipeline (ablation) | BEVFusion · nuScenes val | NDS: 68.71 -> 68.99 | No clear effect |
| large resize augmentation on BEVFusion-like pipeline instead of DAL pipeline (ablation) | BEVFusion · nuScenes val | NDS: 68.99 -> 70.97 | Improved |
| DAL pipeline vs BEVFusion pipeline with velocity augmentation (ablation) | DAL · nuScenes val | NDS: 71.67 -> 71.94 | Improved |
| increase image resolution or image backbone capacity beyond R50 256x704 (ablation) | DAL image branch · nuScenes val | NDS: 71.94 -> 72.13 at 384x1056; 71.94 -> 71.62 with R101 | No clear effect |
| increase LiDAR voxel resolution from 0.100 to 0.075 (ablation) | DAL LiDAR branch · nuScenes val | NDS: 71.94 -> 72.64 | Improved |
| use DAL-Base instead of CMT-R50 at similar speed | DAL-Base · nuScenes val | mAP / NDS: +2.1 mAP and +2.6 NDS | Improved |
| handle objects beyond LiDAR range with a separate fused-feature regression FFN (ablation) | DAL · nuScenes | accuracy | 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.
- UniTR — UniTR: A Unified and Efficient Multi-Modal Transformer for Bird's-Eye-View Representation (ICCV 2023) · 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
- Once Detected, Never Lost: Surpassing Human Performance in Offline LiDAR based 3D Object Detection (ICCV 2023 · similarity 0.91)
- TCC-Det: Temporarily consistent cues for weakly-supervised 3D detection (ECCV 2024 · similarity 0.90)
- Weakly Supervised 3D Object Detection from Lidar Point Cloud (ECCV 2020 · similarity 0.90)
- MixSup: Mixed-grained Supervision for Label-efficient LiDAR-based 3D Object Detection (ICLR 2024 · similarity 0.89)
- STONE: A Submodular Optimization Framework for Active 3D Object Detection (NeurIPS 2024 · similarity 0.89)
- A-Teacher: Asymmetric Network for 3D Semi-Supervised Object Detection (CVPR 2024 · similarity 0.89)
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.
