Research map / Point clouds / Segmentation and detection
EPNet: Enhancing Point Features with Image Semantics for 3D Object Detection
Tengteng Huang, Zhe Liu, Xiwu Chen, Xiang Bai
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 LI-Fusion modules (point-wise image feature fusion) (ablation) | EPNet two-stream RPN · KITTI val 3D | 3D mAP: 79.94 -> 81.67 | Improved |
| add consistency enforcing (CE) loss (ablation) | EPNet · KITTI val 3D | 3D mAP: 79.94 -> 83.87 | Improved |
| add LI-Fusion on top of CE loss (ablation) | EPNet · KITTI val 3D | 3D mAP: 83.87 -> 85.00 | Improved |
| replace CE loss with IoU loss (ablation) | EPNet · KITTI val 3D | 3D mAP: CE loss gives +1.28% 3D mAP over IoU loss | Improved |
| fuse RGB by simple concatenation with point coordinates instead of feature fusion (ablation) | geometric stream · KITTI val 3D | 3D mAP: 79.94 -> 79.66 | Hurt |
| single-scale fusion (last FP layer only) instead of baseline without fusion (ablation) | EPNet · KITTI val 3D | 3D mAP: 79.94 -> 80.36 | Improved |
| use multi-scale LI-Fusion instead of single-scale fusion (ablation) | EPNet · KITTI val 3D | 3D mAP: 80.36 -> 81.67 | Improved |
| simple RGB concatenation under altered illumination (ablation) | LiDAR-only model · KITTI val 3D, altered illumination | 3D mAP: 83.87 -> 82.79 | Hurt |
| LI-Fusion without weight map under altered illumination (ablation) | LiDAR-only model · KITTI val 3D, altered illumination | 3D mAP: 83.87 -> 83.18 | Hurt |
| add LiDAR-guided weight map w to LI-Fusion (ablation) | LI-Fusion · KITTI val 3D, altered illumination | 3D mAP: 83.18 -> 84.52 | Improved |
| add CE loss instead of IoU loss: kept positive candidate box ratio (ablation) | EPNet · KITTI val | ratio of kept positive candidate boxes | Improved |
Later papers that compared with this method
1 later accepted paper reports EPNet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- EPNet++: Cascade Bi-Directional Fusion for Multi-Modal 3D Object Detection (TPAMI 2022 · better on 5 of 5 shared results)
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.
- PointRCNN — PointRCNN: 3D Object Proposal Generation and Detection From Point Cloud (CVPR 2019) · better on 3 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
- EPNet++: Cascade Bi-Directional Fusion for Multi-Modal 3D Object Detection (TPAMI 2022 · similarity 0.92)
- PG-RCNN: Semantic Surface Point Generation for 3D Object Detection (ICCV 2023 · similarity 0.89)
- AGO-Net: Association-Guided 3D Point Cloud Object Detection Network (TPAMI 2021 · similarity 0.88)
- ViKIENet: Towards Efficient 3D Object Detection with Virtual Key Instance Enhanced Network (CVPR 2025 · similarity 0.88)
- Beyond Virtual Points: Depth-Enhanced LiDAR-only 3D Object Detection with Semi-Supervised Learning (Student Abstract) (AAAI 2025 · similarity 0.87)
- Seg2Box: 3D Object Detection by Point-Wise Semantics Supervision (AAAI 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.
