Research map / Point clouds / Point cloud classification
Continuous SO(3) Equivariant Convolution for 3D Point Cloud Analysis
Jaein Kim, HEE BIN YOO, Dong-Sig Han, Yeon-Ji Song, Byoung-Tak Zhang
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 continuous SO(3) equivariant filters instead of discrete group convolution when test rotations differ from training | CSEConv vs E2PN · ModelNet40, train I / test SO(3) | accuracy (%): 44.47 -> 83.75 | Improved |
| apply SO(3) augmentation to non-equivariant model training instead of none | PCT · ModelNet40, test SO(3) | accuracy (%): 15.84 -> 84.20 | Improved |
| add SO(3) augmentation during training for aligned test set | non-equivariant PCT · ModelNet40, test I | accuracy (%): 91.25 -> 86.39 | Hurt |
| use continuous-group equivariant filters instead of discretized group | equivariance to random rotations · ModelNet40, no train augmentation | accuracy (%): 44.47 -> 83.75 | Improved |
| double the number of KNN neighbors in the integration (ablation) | CSEConv · equivariance error on 512 points | equivariance error | Improved |
| implement filter as small weight-sharing coordinate network instead of tensor grid parameters | CSEConv · ModelNet40 training speed | batches per second: 8.64 -> 27.46 | Improved |
| use CSEConv instead of E2PN on real-world scans with aligned train and test | ScanObjectNN OBJ_ONLY · ScanObjectNN, train I / test I | accuracy (%): 82.70 -> 72.29 | Hurt |
| use CSEConv instead of Frame Averaging, SE3-T and VN on ScanObjectNN | continuous group equivariant methods · ScanObjectNN, train I / test SO(3) | accuracy (%): 71.94 -> 72.29 (vs FA) | Improved |
Closest papers by meaning
- E2PN: Efficient SE(3)-Equivariant Point Network (CVPR 2023 · similarity 0.93)
- Fast, Expressive $\mathrm{SE}(n)$ Equivariant Networks through Weight-Sharing in Position-Orientation Space (ICLR 2024 · similarity 0.88)
- SIEVE: Overcoming Topological Obstruction in Equivariant Self-Supervised Learning (NeurIPS 2026 · similarity 0.88)
- Towards Learning Group-Equivariant Features for Domain Adaptive 3D Detection (NeurIPS 2024 · similarity 0.87)
- Approximately Piecewise E(3) Equivariant Point Networks (ICLR 2024 · similarity 0.86)
- Probing Equivariance and Symmetry Breaking in Convolutional Networks (NeurIPS 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.
