Research map / Point clouds / Point cloud classification
LPD-Net: 3D Point Cloud Learning for Large-Scale Place Recognition and Environment Analysis
Zhe Liu, Shunbo Zhou, Chuanzhe Suo, Peng Yin, Wen Chen, Hesheng Wang, Haoang Li, Yun-Hui 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 |
|---|---|---|---|
| add Feature Network adaptive local features to the xyz-only input (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 84.74 -> 94.92 | Improved |
| add F2D local features (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 90.76 -> 94.92 | Improved |
| add F3D local features (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 91.23 -> 94.92 | Improved |
| add four extra F3D features beyond the ten selected (14 total) (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 94.92 -> 92.03 | Hurt |
| add graph-based neighborhood aggregation (series Feature+Cartesian) to FN-VLAD (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 89.77 -> 94.92 | Improved |
| use series-FC aggregation instead of parallel-maxpooling (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 91.20 -> 94.92 | Improved |
| use series-FC aggregation instead of parallel-concatenation (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 92.27 -> 94.92 | Improved |
| use parallel-concatenation instead of parallel-maxpooling (ablation) | FN-PM-VLAD · Oxford RobotCar place recognition | Ave recall @1%: 91.20 -> 92.27 | Improved |
| use parallel feature-relation structure instead of series (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 92.60 -> 94.92 | Improved |
| use parallel feature-relation structure instead of original (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 91.53 -> 94.92 | Improved |
| replace adaptive local features with learned kNN+mlp features (NN-VLAD) (ablation) | FN-SF-VLAD · Oxford RobotCar place recognition | Ave recall @1%: 94.92 -> 79.21 | Hurt |
| use K=64 instead of K=128 visual words at D=256 (ablation) | LPD-Net (FN-SF-VLAD) · Oxford RobotCar place recognition | Ave recall @1%: 92.47 -> 94.92 | Improved |
Closest papers by meaning
- PointNetVLAD: Deep Point Cloud Based Retrieval for Large-Scale Place Recognition (CVPR 2018 · similarity 0.91)
- Pyramid Point Cloud Transformer for Large-Scale Place Recognition (ICCV 2021 · similarity 0.91)
- PCAN: 3D Attention Map Learning Using Contextual Information for Point Cloud Based Retrieval (CVPR 2019 · similarity 0.88)
- SOE-Net: A Self-Attention and Orientation Encoding Network for Point Cloud Based Place Recognition (CVPR 2021 · similarity 0.87)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.86)
- DeepLA-Net: Very Deep Local Aggregation Networks for Point Cloud Analysis (CVPR 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.
