Research map / Architectures and efficiency / CNN design
Learning Steerable Filters for Rotation Equivariant 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 coefficient-aware initialization (CoeffInit) instead of He initialization (ablation) | SFCNN · rotated MNIST | test error (%): 0.957 ± 0.025 -> 0.880 ± 0.029 | Helped |
| Add continuous train-time rotations (ablation) | SFCNN with CoeffInit · rotated MNIST | test error (%): 0.880 ± 0.029 -> 0.714 ± 0.022 | Helped |
| Increase sampled filter orientations from 2 to 4 (ablation) | SFCNN · rotated MNIST | test error | Helped |
| Increase sampled orientations beyond four, up to around 12 to 16 (ablation) | SFCNN · rotated MNIST | test error | Helped |
| Add augmentation by rotations in a subgroup under which the SFCNN is equivariant (ablation) | SFCNN · ISBI 2012 EM segmentation | segmentation performance | No clear effect |
| Use a conventional CNN rather than an SFCNN | rotational generalization experiment · rotated MNIST | test error across rotated test images | Hurt |
| Increase the filter-mask size (ablation) | SFCNN · rotated MNIST | test error | Helped |
Closest papers by meaning
- RotDCF: Decomposition of Convolutional Filters for Rotation-Equivariant Deep Networks (ICLR 2019 · similarity 0.92)
- Steerable CNNs (ICLR 2017 · similarity 0.91)
- A Probabilistic Approach to Learning the Degree of Equivariance in Steerable CNNs (ICML 2024 · similarity 0.89)
- Steerable Transformers for Volumetric Data (ICML 2025 · similarity 0.87)
- Rotation-Equivariant Self-Supervised Method in Image Denoising (CVPR 2025 · similarity 0.87)
- Clifford-Steerable Convolutional Neural Networks (ICML 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.