Research map / Point clouds / Point cloud classification
A Closer Look at Rotation-Invariant Deep Point Cloud Analysis
Feiran Li, Kent Fujiwara, Fumio Okura, Yasuyuki Matsushita
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 the pose selector module over the 24 canonical poses (ablation) | DGCNN with PCA canonical poses · ModelNet40 | accuracy (%): 91.3 -> 91.6 | Improved |
| add the pose selector module over the 24 canonical poses (ablation) | DGCNN with PCA canonical poses · ShapeNet part | mIoU (%): 82.8 -> 83.1 | Improved |
| use all 24 ambiguities (4 sign x 6 order) instead of fixing order by eigenvalue sorting (Strategy I, 4 ambiguities) (ablation) | DGCNN with PCA canonical poses · ModelNet40 | accuracy (%): 89.5 -> 91.6 | Improved |
| use all 24 ambiguities instead of fixing signs by majority of data (Strategy II, 6 ambiguities) (ablation) | DGCNN with PCA canonical poses · ModelNet40 | accuracy (%): 90.3 -> 91.6 | Improved |
| use all 24 ambiguities instead of eliminating all ambiguities (Strategy III) (ablation) | DGCNN with PCA canonical poses · ModelNet40 | accuracy (%): 87.2 -> 91.6 | Improved |
| use 8 sign combinations (including improper rotations) instead of the correct ambiguities (ablation) | DGCNN with PCA canonical poses · ModelNet40 | accuracy (%): 89.2%, 2.4% lower than with correctly identified ambiguities; worse than Strategy I (89.5) | Hurt |
| manually pick the best of 24 poses instead of feeding all 24 poses (ablation) | DGCNN with PCA canonical poses · ModelNet40 | accuracy (%): 91.6 -> 92.0 | Improved |
| manually pick the best of 24 poses instead of feeding all 24 poses (ablation) | DGCNN with PCA canonical poses · ShapeNet part | mIoU (%): 83.1 -> 84.7 | Improved |
| use manually pre-aligned pose instead of the selected PCA canonical pose | DGCNN · ModelNet40 | accuracy (%): 92.0 -> 92.9 | Improved |
| add test-time augmentation over 24 concatenation patterns | pose selector DGCNN · ModelNet40 | accuracy (%): 90.2 -> 91.6 | Improved |
| add test-time augmentation over 24 concatenation patterns | pose selector DGCNN · ScanObjectNN OBJ_BG | accuracy (%): 84.3 -> 86.7 | Improved |
| use correctly identified PCA ambiguities with pose selector instead of LGR-Net | real-world rotation-invariant classification · ScanObjectNN OBJ_BG | accuracy (%): 81.4 -> 86.7 (5.3% enhancement) | Improved |
Closest papers by meaning
- Deep Positional and Relational Feature Learning for Rotation-Invariant Point Cloud Analysis (ECCV 2020 · similarity 0.88)
- Rethinking Rotation Invariance with Point Cloud Registration (AAAI 2023 · similarity 0.87)
- RIConv++: Effective Rotation Invariant Convolutions for 3D Point Clouds Deep Learning (IJCV 2022 · similarity 0.87)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.85)
- Local-consistent Transformation Learning for Rotation-invariant Point Cloud Analysis (CVPR 2024 · similarity 0.85)
- CRA-PCN: Point Cloud Completion with Intra- and Inter-level Cross-Resolution Transformers (AAAI 2024 · similarity 0.84)
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.
