Research map / Point clouds / Point cloud classification
PPF-FoldNet: Unsupervised Learning of Rotation Invariant 3D Local Descriptors
Haowen Deng, Tolga Birdal, Slobodan Ilic
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 |
|---|---|---|---|
| use 5K points per local patch instead of 2K (ablation) | PPF-FoldNet · 3DMatch (standard) | recall: 0.6804 -> 0.7182 | Improved |
| use 5K points per local patch instead of 2K (ablation) | PPF-FoldNet · Rotated 3DMatch | recall: 0.6865 -> 0.7311 | Improved |
| replace PPFNet with PPF-FoldNet (unsupervised, PPF only) | PPFNet · 3DMatch (standard) | recall: 62.32% -> 68.04% | Improved |
| replace FPFH with PPF-FoldNet to compress PPFs | FPFH · 3DMatch (standard) | recall: 0.3589 -> 0.6804 | Improved |
| use PPF-FoldNet instead of PPFNet under random rotations | PPFNet · Rotated 3DMatch | recall: 0.0026 -> 0.6865 | Improved |
| use PPF-FoldNet instead of CGF under random rotations | CGF · Rotated 3DMatch | recall: 0.4987 -> 0.6865 | Improved |
| use Bobkov1 PPF (with occupancy ratio) instead of the proposed PPF (ablation) | PPF-FoldNet · Rotated 3DMatch subset | recall: 0.463 -> 0.483 | No clear effect |
| use PPFH (FPFH-style PPF) instead of the proposed PPF (ablation) | PPF-FoldNet · Rotated 3DMatch subset | recall: 0.463 -> 0.436 | No clear effect |
| use PPF-FoldNet instead of FPFH for feature computation | FPFH · 3DMatch fragment | time (s): 31.678 -> 3.969 | Improved |
| decrease fragment points from 100% to 6.25% (ablation) | PPF-FoldNet · 3DMatch sparsity test | recall: greater than 50% | No clear effect |
| decrease fragment points to 6.25% for PPFNet (ablation) | PPFNet · 3DMatch sparsity test | recall: around 12% | Hurt |
| compare against PPFNet on standard scenes Kitchen | PPF-FoldNet (2K) · 3DMatch Kitchen | recall: 0.8972 -> 0.7352 | Hurt |
Later papers that compared with this method
2 later accepted papers report PPF-FoldNet-5K as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- 3D Point Capsule Networks (CVPR 2019 · better on 2 of 2 shared results)
- 3D Local Features for Direct Pairwise Registration (CVPR 2019 · better on 1 of 1 shared result)
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.
- PPFNet — PPFNet: Global Context Aware Local Features for Robust 3D Point Matching (CVPR 2018) · 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
- PPFNet: Global Context Aware Local Features for Robust 3D Point Matching (CVPR 2018 · similarity 0.91)
- Unsupervised Learning of Local Equivariant Descriptors for Point Clouds (TPAMI 2021 · similarity 0.90)
- RIGA: Rotation-Invariant and Globally-Aware Descriptors for Point Cloud Registration (TPAMI 2024 · similarity 0.89)
- Rotation-Invariant Local-to-Global Representation Learning for 3D Point Cloud (NeurIPS 2020 · similarity 0.88)
- You Only Train Once: Learning General and Distinctive 3D Local Descriptors (TPAMI 2022 · similarity 0.88)
- Enhancing Rotation-Invariant 3D Learning with Global Pose Awareness and Attention Mechanisms (AAAI 2026 · similarity 0.88)
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.
