Research map / Architectures and efficiency / CNN design
Equivariant Networks for Pixelized Spheres
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 symmetry-constrained global operations rather than gauge-equivariant operations | Equivariant networks for pixelized spheres · Spherical MNIST (tetrahedron, cube, octahedron, icosahedron) | Accuracy (%): 98.37 ± 0.07 -> 98.73 ± 0.08; 98.70 ± 0.06 -> 98.99 ± 0.05; 98.25 ± 0.11 -> 99.02 ± 0.05; 97.45 ± 0.12 -> 98.93 ± 0.11 | Helped |
| Constrain face transformations to be determined by the solid's symmetry instead of independent face transformations | Equivariant networks for pixelized spheres · Spherical MNIST (tetrahedron, cube, octahedron, icosahedron) | Accuracy (%): 98.62 ± 0.10 -> 98.73 ± 0.08; 98.82 ± 0.03 -> 98.99 ± 0.05; 98.62 ± 0.03 -> 99.02 ± 0.05; 98.54 ± 0.05 -> 98.93 ± 0.11 | Helped |
| Use the orientation-preserving subgroup for omnidirectional images | Equivariant segmentation network · Stanford 2D3DS | Mean Acc. (%): 58.74 -> 62.5 | Helped |
| Use the orientation-preserving subgroup for omnidirectional images | Equivariant segmentation network · Stanford 2D3DS | Mean IoU (%): 40.99 -> 45.0 | Helped |
| Use the proposed equivariant network | Cube pixelization segmentation network · HAPPI20 | Mean Acc. (%): 97.8 ± 0.3 -> 99.30 | Helped |
| Use the proposed equivariant network | Cube pixelization segmentation network · HAPPI20 | Mean AP (%): 80.6 -> 95.20 | Helped |
| Use 25% rather than 100% of channels for global operations (ablation) | Final equivariant model · Spherical MNIST (tetrahedron) | Accuracy (%): 98.72 ± 0.09 -> 98.79 ± 0.03 | No clear effect |
| Use 25% rather than 100% of channels for global operations (ablation) | Final equivariant model · Spherical MNIST (cube) | Accuracy (%): 98.83 ± 0.05 -> 98.94 ± 0.09 | No clear effect |
| Use 25% rather than 100% of channels for global operations (ablation) | Final equivariant model · Spherical MNIST (octahedron) | Accuracy (%): 98.95 ± 0.07 -> 98.96 ± 0.08 | No clear effect |
| Use 25% rather than 100% of channels for global operations (ablation) | Final equivariant model · Spherical MNIST (icosahedron) | Accuracy (%): 98.84 ± 0.06 -> 98.82 ± 0.07 | No clear effect |
| Use the oriented model rather than the non-oriented model (ablation) | Equivariant climate segmentation network · HAPPI20 | Mean accuracy | No clear effect |
Closest papers by meaning
- Gauge Equivariant Convolutional Networks and the Icosahedral CNN (ICML 2019 · similarity 0.93)
- AtlasD: Automatic Local Symmetry Discovery (ICML 2025 · similarity 0.86)
- Probing Equivariance and Symmetry Breaking in Convolutional Networks (NeurIPS 2025 · similarity 0.86)
- On the Fourier analysis in the SO(3) space : the EquiLoPO Network (ICLR 2025 · similarity 0.86)
- Fast, Expressive $\mathrm{SE}(n)$ Equivariant Networks through Weight-Sharing in Position-Orientation Space (ICLR 2024 · similarity 0.86)
- Approximately Piecewise E(3) Equivariant Point Networks (ICLR 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.