Research map / Architectures and efficiency / CNN design
Vector Neurons: A General Framework for SO(3)-Equivariant Networks
Congyue Deng, Or Litany, Yueqi Duan, Adrien Poulenard, Andrea Tagliasacchi, Leonidas J. Guibas
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 PointNet with VN-PointNet | PointNet · ModelNet40 (z/SO(3)) | accuracy: 19.6 -> 77.5 | Improved |
| replace DGCNN with VN-DGCNN | DGCNN · ModelNet40 (z/SO(3)) | accuracy: 33.8 -> 89.5 | Improved |
| replace DGCNN with VN-DGCNN under full rotation augmentation | DGCNN · ModelNet40 (SO(3)/SO(3)) | accuracy: 88.6 -> 90.2 | Improved |
| replace DGCNN with VN-DGCNN | DGCNN · ShapeNet-part (z/SO(3)) | mIoU: 49.3 -> 81.4 | Improved |
| train on aligned data, test on arbitrary rotations: VN-DGCNN vs DGCNN | DGCNN · ModelNet40 (I/SO(3)) | accuracy: 16.6 -> 90.0 | Improved |
| train on aligned data, test on arbitrary rotations: VN-PointNet vs PointNet | PointNet · ShapeNet-part (I/SO(3)) | mIoU: 30.3 -> 73.0 | Improved |
| use equivariant VN-PointNet encoder in OccNet | OccNet · ShapeNet reconstruction (SO(3)/SO(3)) | volumetric mIoU: 58.2 -> 68.8 | Improved |
| use equivariant VN-PointNet encoder in OccNet on aligned data | OccNet · ShapeNet reconstruction (I/I) | volumetric mIoU: 71.4 -> 69.3 | Hurt |
| add invariant decoder with standard PointNet encoder (ablation) | OccNet · ShapeNet reconstruction (I/I) | volumetric mIoU: 71.4 -> 72.0 | Improved |
| use detached linear and non-linear layers in VN-DGCNN (ablation) | VN-DGCNN · ModelNet40 (z/z) | accuracy: 89.5 -> 90.8 | Improved |
| use detached linear and non-linear layers in VN-PointNet (ablation) | VN-PointNet · ModelNet40 | accuracy: 77.5 -> 78.2 (z/z); 77.2 -> 76.8 (SO(3)/SO(3)) | Mixed |
| use mean pooling instead of VN-MAX pooling in VN-DGCNN (ablation) | VN-DGCNN · ModelNet40 (z/z) | accuracy: 88.9 -> 89.5 | Improved |
Closest papers by meaning
- An Intuitive Multi-Frequency Feature Representation for SO(3)-Equivariant Networks (ICLR 2024 · similarity 0.91)
- Reducing SO(3) Convolutions to SO(2) for Efficient Equivariant GNNs (ICML 2023 · similarity 0.90)
- A Functional Approach to Rotation Equivariant Non-Linearities for Tensor Field Networks. (CVPR 2021 · similarity 0.90)
- Ada-VNNs: Adaptive Equivariance for Vector Neural Networks (ECCV 2026 · similarity 0.90)
- A Characterization Theorem for Equivariant Networks with Point-wise Activations (ICLR 2024 · similarity 0.90)
- A Tale of Two Symmetries: Exploring the Loss Landscape of Equivariant Models (NeurIPS 2025 · 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.
