Research map / Point clouds / Point cloud classification
DH3D: Deep Hierarchical 3D Descriptors for Robust Large-Scale 6DoF Relocalization
Juan Du, Rui Wang, Daniel Cremers
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 self-supervision via synthesized transformations instead of weak submap-level supervision (ablation) | DH3D local part · Oxford RobotCar registration | registration performance | Improved |
| add the keypoint detector confidence map during training (ablation) | DH3D local descriptor · Oxford RobotCar registration | registration performance | Improved |
| add SE blocks to the local encoder (ablation) | DH3D local descriptor · Oxford RobotCar registration | registration performance | Improved |
| describe-and-detect rather than detect-then-describe | DH3D detector · Oxford RobotCar repeatability, 256 keypoints | relative repeatability: roughly 1.75x than the best follower 3DFeatNet | Improved |
| use DH3D learned detector instead of random keypoints | DH3D descriptor · Oxford RobotCar registration | success rate (%): 92.1 -> 98.5 | Improved |
| move from LiDAR point clouds to Stereo DSO point clouds without fine-tuning | 3DFeatNet + 3DFeatNet · Stereo DSO registration | success rate (%): 98.1 -> 84.1 | Hurt |
| move from LiDAR point clouds to Stereo DSO point clouds without fine-tuning | USIP + 3DSmoothNet · Stereo DSO registration | success rate (%): 98.0 -> 77.3 | Hurt |
| move from LiDAR point clouds to Stereo DSO point clouds without fine-tuning | USIP + 3DFeatNet · Stereo DSO registration | success rate (%): 99.1 -> 69.9 | Hurt |
| move from LiDAR point clouds to Stereo DSO point clouds without fine-tuning | DH3D detector + DH3D descriptor · Stereo DSO registration | success rate (%): 98.5 -> 90.6 | Hurt |
| halve the points entering NetVLAD (4096 instead of 8192) | DH3D global descriptor · Oxford RobotCar retrieval | average recall | No clear effect |
| add Gaussian noise above 0.2m | DH3D registration · Oxford RobotCar registration | success rate: 0.15 | Hurt |
Closest papers by meaning
- Learning general and distinctive 3D local deep descriptors for point cloud registration (TPAMI 2022 · similarity 0.91)
- D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features (CVPR 2020 · similarity 0.90)
- PPFNet: Global Context Aware Local Features for Robust 3D Point Matching (CVPR 2018 · similarity 0.90)
- End-to-End Learning Local Multi-View Descriptors for 3D Point Clouds (CVPR 2020 · similarity 0.89)
- RIGA: Rotation-Invariant and Globally-Aware Descriptors for Point Cloud Registration (TPAMI 2024 · similarity 0.89)
- PointCFormer: A Relation-Based Progressive Feature Extraction Network for Point Cloud Completion (AAAI 2025 · similarity 0.87)
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.
