Research map / Architectures and efficiency / CNN design
Spin-Weighted Spherical CNNs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 spin-weighted convolutions instead of isotropic spherical convolutions | Spherical CNN · Spherical MNIST NR/R | accuracy: 98.08 +- 0.24 -> 99.08 +- 0.12 | Helped |
| Use spin-weighted vector-field equivariance | Spherical vector field classifier · SVF-MNIST NR/R | accuracy: 24.8 +- 0.8 -> 98.2 +- 0.7 | Helped |
| Use spin-weighted convolutions rather than planar convolutions | Image-to-vector-field predictor · SVF-MNIST easy R/R | MSE x 10^3: 5.0 +- 0.1 -> 3.4 +- 0.1 | Helped |
| Use spin-weighted convolutions rather than planar convolutions | Image-to-vector-field predictor · SVF-MNIST easy NR/NR | MSE x 10^3: 0.3 +- 0.1 -> 2.9 +- 0.2 | Hurt |
| Use spin-weighted convolutions rather than planar convolutions | Vector-field-to-image predictor · SVF-MNIST hard R/R | MSE x 10^3: 13.4 +- 0.2 -> 2.7 +- 0.1 | Helped |
| Use spin-weighted convolutions rather than planar convolutions | Vector-field-to-image predictor · SVF-MNIST easy NR/NR | MSE x 10^3: 1.4 +- 0.1 -> 3.5 +- 0.1 | Hurt |
| Break equivariance in the final layers (ablation) | Ours on ModelNet40 · ModelNet40 upright | accuracy: 89.6 +- 0.3 -> 90.1 +- 0.3 | Helped |
| Break equivariance in the final layers (ablation) | Ours on ModelNet40 · ModelNet40 rotated | accuracy: 88.8 +- 0.1 -> 88.2 +- 0.2 | Hurt |
| Add surface normals (ablation) | Ours on Stanford 2D3DS · Stanford 2D3DS | mIoU: 41.9(5) -> 43.4(4) | Helped |
| Add surface normals and break equivariance in the final layer (ablation) | Ours on Stanford 2D3DS · Stanford 2D3DS | pixel accuracy: 55.6(5) -> 58.7(5) | Helped |
Closest papers by meaning
- Efficient Generalized Spherical CNNs (ICLR 2021 · similarity 0.92)
- Scalable and Equivariant Spherical CNNs by Discrete-Continuous (DISCO) Convolutions (ICLR 2023 · similarity 0.91)
- Rotation Equivariant Graph Convolutional Network for Spherical Image Classification (CVPR 2020 · similarity 0.91)
- Attention on the Sphere (NeurIPS 2025 · similarity 0.87)
- Clifford-Steerable Convolutional Neural Networks (ICML 2024 · similarity 0.86)
- O$n$ Learning Deep O($n$)-Equivariant Hyperspheres (ICML 2024 · 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.