Research map / Architectures and efficiency / CNN design
Scaling-Translation-Equivariant Networks with Decomposed Convolutional Filters
Wei Zhu, Qiang Qiu, Robert Calderbank, Guillermo Sapiro, Xiuyuan Cheng
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 |
|---|---|---|---|
| use ScDCFNet instead of SESN | ScDCFNet · STL-10 | accuracy (%): 83.89 -> 84.90 | Improved |
| use ScDCFNet instead of SESN, 2K train | ScDCFNet · SMNIST 28x28 no augmentation | accuracy (%): 95.75 -> 95.87 | Improved |
| use ScDCFNet instead of SESN, 2K train | ScDCFNet · SFashion 56x56 no augmentation | accuracy (%): 82.34 -> 83.01 | Improved |
| use ScDCFNet instead of regular CNN, 2K train | ScDCFNet · SFashion 28x28 no augmentation | accuracy (%): 79.88 -> 81.32 | Improved |
| add scale data augmentation to ScDCFNet, 2K train | ScDCFNet · SMNIST 28x28 | accuracy (%): 95.87 -> 96.68 | Improved |
| reduce spatial basis from K=15 to K=6 (add more basis functions) (ablation) | ScDCFNet · SMNIST 28x28 2K | accuracy (%): 95.61 -> 95.87 | Improved |
| reduce spatial basis from K=15 to K=6 (add more basis functions) (ablation) | ScDCFNet · SFashion 28x28 2K | accuracy (%): 80.88 -> 81.32 | Improved |
| increase K from 12 to 15 (ablation) | ScDCFNet · SMNIST 28x28 2K | accuracy (%): 95.88 -> 95.87 | No clear effect |
| increase scale basis Kα from 1 to 3 at K=15 (ablation) | ScDCFNet · SFashion 28x28 2K | accuracy (%): 80.74 -> 81.32 | Improved |
| use one-sided-replicate-padding in scale instead of zero-padding | ScDCFNet · equivariance error vs depth | relative L2 equivariance error | Improved |
| use smaller scale filter support Iα | ScDCFNet · equivariance error vs depth | relative L2 equivariance error | Improved |
| reduce spatial basis K in auto-encoder (K=5,8,12,15) (ablation) | ScDCFNet auto-encoder · SMNIST reconstruction | relative L2 reconstruction error | Improved |
Closest papers by meaning
- Scale-Equivariant Steerable Networks (ICLR 2020 · similarity 0.93)
- Empowering Networks With Scale and Rotation Equivariance Using A Similarity Convolution (ICLR 2023 · similarity 0.92)
- A Probabilistic Approach to Learning the Degree of Equivariance in Steerable CNNs (ICML 2024 · similarity 0.87)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.86)
- Learning Color Equivariant Representations (ICLR 2025 · similarity 0.86)
- Normalize Filters! Classical Wisdom for Deep Vision (NeurIPS 2025 · 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.
