Research map / Point clouds / Point cloud classification
Revisiting Transformation Invariant Geometric Deep Learning: An Initial Representation Perspective
Ziwei Zhang, Xin Yue Wang, Zeyang Zhang, Peng Cui, Wenwu Zhu
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 |
|---|---|---|---|
| replace raw coordinates with TinvNet initial representation | PointNet · ModelNet40 z/SO3 | accuracy (%): 13.4 -> 86.2 | Improved |
| replace raw coordinates with TinvNet initial representation | DGCNN · ModelNet40 z/SO3 | accuracy (%): 22.3 -> 89.5 | Improved |
| replace raw coordinates with TinvNet representation | DGCNN · ModelNet40 z/z | accuracy (%): 92.2 -> 89.5 | Hurt |
| replace raw coordinates with TinvNet representation | PointNet · ShapeNet z/SO3 | mean IoU (%): 29.0 -> 80.0 | Improved |
| add TinvNet to GAT | GAT · TSP-100 translation | route length: 52.15 -> 8.52 | Improved |
| add TinvNet to GAT | GAT · CVRP-100 translation | route length: 104.23 -> 17.31 | Improved |
| add TinvNet to GAT | GAT · TSP-100 no transformation | route length: 8.49 -> 8.52 | No clear effect |
| add TinvNet to GAT | GAT · CVRP-100 no transformation | route length: 17.73 -> 17.31 | Improved |
| replace TinvNet representation with k-NN distances (ablation) | PointNet · ModelNet40 z/SO3 | accuracy (%): 86.2 -> 24.9 | Hurt |
| replace TinvNet representation with PCA whitening (ablation) | DGCNN · ModelNet40 z/SO3 | accuracy (%): 89.5 -> 85.8 | Hurt |
| enumerate 2^d eigenvector signs as augmentation instead of canonical sign (ablation) | TinvNet(PointNet) · ModelNet40 z/SO3 | accuracy (%): 85.4 -> 86.2 | Improved |
| enumerate 2^d eigenvector signs instead of canonical sign (ablation) | TinvNet(DGCNN) · ModelNet40 z/z | accuracy (%): 88.3 -> 89.5 | Improved |
Closest papers by meaning
- Graph-based Isometry Invariant Representation Learning (ICML 2017 · similarity 0.91)
- On the Completeness of Invariant Geometric Deep Learning Models (ICLR 2025 · similarity 0.89)
- A Characterization Theorem for Equivariant Networks with Point-wise Activations (ICLR 2024 · similarity 0.87)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.86)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · similarity 0.86)
- Geometric Inductive Biases of Deep Networks: The Role of Data and Architecture (ICLR 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.
