Research map / Architectures and efficiency / CNN design
Equivariance versus Augmentation for Spherical Images
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 rotational data augmentation and a larger non-equivariant CNN instead of a smaller equivariant network | Spherical MNIST classification · Spherical MNIST | accuracy: 97.64% -> 97.49% | No clear effect |
| Increase rotational data augmentation for a non-equivariant CNN | Spherical MNIST classification · Spherical MNIST | accuracy | Helped |
| Use rotational data augmentation rather than built-in rotational equivariance | Spherical MNIST semantic segmentation · One-digit spherical MNIST | non-background mIoU | Hurt |
| Increase augmented training data for non-equivariant CNNs | Spherical MNIST semantic segmentation · One-digit spherical MNIST | non-background mIoU | No clear effect |
| Increase the S2CNN parameter count from 204k to 820k | S2CNN · Spherical MNIST semantic segmentation, one digit, 60k training samples | non-background mIoU: 78.28% -> 76.71% | Hurt |
| Increase task complexity by projecting four digits per sphere | Equivariant versus non-equivariant segmentation models · Four-digit spherical MNIST | non-background mIoU | Helped |
| Replace MNIST digits with FashionMNIST clothing items | Equivariant versus non-equivariant segmentation models · Spherical FashionMNIST | non-background mIoU | Mixed |
| Train a non-equivariant CNN on unrotated data, then evaluate on rotated data | Non-equivariant segmentation models · Spherical MNIST segmentation | non-background mIoU | Hurt |
| Project unrotated digits at the pole instead of the grid center | Non-equivariant segmentation models · Spherical MNIST segmentation | non-background mIoU | Mixed |
| Add equivariant skip connections (ablation) | S2CNN segmentation model · Spherical MNIST semantic segmentation | non-background mIoU | No clear effect |
Closest papers by meaning
- Scalable and Equivariant Spherical CNNs by Discrete-Continuous (DISCO) Convolutions (ICLR 2023 · similarity 0.92)
- Efficient Generalized Spherical CNNs (ICLR 2021 · similarity 0.91)
- Empowering Networks With Scale and Rotation Equivariance Using A Similarity Convolution (ICLR 2023 · similarity 0.91)
- A Probabilistic Approach to Learning the Degree of Equivariance in Steerable CNNs (ICML 2024 · similarity 0.88)
- Probing Equivariance and Symmetry Breaking in Convolutional Networks (NeurIPS 2025 · similarity 0.88)
- Equivariant Neural Tangent Kernels (ICML 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.