Research map / 3D vision / Human pose estimation
Ada-VNNs: Adaptive Equivariance for Vector Neural Networks
Bing Han, Ruitao Pan, Yumin Chen, Peixin Hong, Weiyuan Liu, Zhibin Zhao, Chenxi Wang, Zhi Zhai
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 Adaptive Gating Unit and residual pathway to VNN | VNN · ShapeNet continuous self-symmetry (avg) | 10° registration recall: 8.87 -> 70.31 | Improved |
| add the Adaptive Gating Unit and residual pathway to VNN | VNN · ShapeNet discrete self-symmetry (avg) | 10° registration recall: 21.29 -> 70.41 | Improved |
| add the Adaptive Gating Unit and residual pathway to VNN | VNN · ShapeNet asymmetric (avg) | 10° registration recall: 67.19 -> 86.24 | Improved |
| add the Adaptive Gating Unit and residual pathway to VN-Transformer | VN-T · ShapeNet continuous self-symmetry (avg) | 10° registration recall: 51.17 -> 72.09 | Improved |
| add the Adaptive Gating Unit and residual pathway to VN-Transformer | VN-T · ShapeNet discrete self-symmetry (avg) | 10° registration recall: 76.10 -> 80.49 | Improved |
| add the Adaptive Gating Unit and residual pathway to VN-Transformer | VN-T · ShapeNet asymmetric (avg) | 10° registration recall: 86.10 -> 85.77 | No clear effect |
| add the Adaptive Gating Unit and residual pathway to VNN | VNN · Isotope prediction (C6H6) | angular error (rad): 0.946 -> 0.302 | Improved |
| add the adaptive branch to VN-DGCNN | VN-DGCNN · ModelNet40 z/SO(3) | accuracy: 90.2 -> 86.9 | Hurt |
| add the adaptive branch to VN-PointNet | VN-PointNet · ModelNet40 z/z | accuracy: 77.5 -> 81.2 | Improved |
| add the adaptive branch to VN-DGCNN | VN-DGCNN · ModelNet40 invariance error | invariance error: 0.0023 -> 0.0040 | Hurt |
| increase AGLL layers in AGU from 1 to 2 (ablation) | Ada-VNN · ShapeNet Lamp | 10° registration recall: ~30 -> 79.44 | Improved |
| increase AGLL layers k in AGU (ablation) | Ada-VNN · ShapeNet 26 categories | equivariance error: roughly one order of magnitude per added layer | Hurt |
Closest papers by meaning
- Learning (Approximately) Equivariant Networks via Constrained Optimization (NeurIPS 2025 · similarity 0.89)
- Bridging Symmetry and Robustness: On the Role of Equivariance in Enhancing Adversarial Robustness (NeurIPS 2025 · similarity 0.88)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.88)
- A Probabilistic Approach to Learning the Degree of Equivariance in Steerable CNNs (ICML 2024 · similarity 0.88)
- A Tale of Two Symmetries: Exploring the Loss Landscape of Equivariant Models (NeurIPS 2025 · similarity 0.87)
- Improving Equivariant Model Training via Constraint Relaxation (NeurIPS 2024 · 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.
