Research map / Architectures and efficiency / CNN design
RotDCF: Decomposition of Convolutional Filters for Rotation-Equivariant Deep Networks
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 |
|---|---|---|---|
| Replace regular convolution with rotation-equivariant RotDCF layers | Conv-3 CNN · rotMNIST, 10K training | test accuracy: 95.67 -> 97.59 | Helped |
| Replace regular convolution with rotation-equivariant RotDCF layers | Conv-3 CNN · rotMNIST, 5K training | test accuracy: 94.04 -> 96.79 | Helped |
| Replace regular convolution with RotDCF layers using K=3, K_alpha=5 | VGG-16 CNN · CIFAR10, 10K training | test accuracy: 78.40 -> 79.53 | Helped |
| Train with rotation-equivariant RotDCF instead of CNN | MNIST classifier transferred to rotated MNIST · MaxRot=30 degrees, no-retrain | test accuracy: 92.61 -> 96.90 | Helped |
| Train with rotation-equivariant RotDCF instead of CNN | MNIST classifier transferred to rotated MNIST · MaxRot=30 degrees, fc-retrain | test accuracy: 94.71 -> 98.48 | Helped |
| Train with rotation-equivariant RotDCF instead of CNN | MNIST classifier transferred to rotated MNIST · MaxRot=60 degrees, no-retrain | test accuracy: 69.61 -> 82.36 | Helped |
| Train with rotation-equivariant RotDCF instead of CNN | MNIST classifier transferred to rotated MNIST · MaxRot=60 degrees, fc-retrain | test accuracy: 85.90 -> 97.68 | Helped |
| Use RotDCF instead of CNN, with circular feature alignment | Facescrub face recognition · Known subjects, in-plane rotation | recognition accuracy: 0.54% -> 97.04% | Helped |
| Use RotDCF instead of CNN | Facescrub face recognition · Known subjects, out-of-plane rotation | recognition accuracy: 80.79% -> 89.66% | Helped |
| Use RotDCF instead of CNN | Facescrub face recognition · Unknown subjects, upright | recognition accuracy: 96.92% -> 96.92% | No clear effect |
| Use RotDCF instead of CNN | Facescrub face recognition · Known subjects, upright | recognition accuracy: 94.10% -> 93.42% | Hurt |
| Reduce spatial basis count from K=14 to K=5 (ablation) | RotDCF, M=16, K_alpha=8 · rotMNIST, 10K training | test accuracy: 97.86 -> 97.81 | No clear effect |
Closest papers by meaning
- Learning Steerable Filters for Rotation Equivariant CNNs (CVPR 2018 · similarity 0.92)
- General E(2)-Equivariant Steerable CNNs (NeurIPS 2019 · similarity 0.90)
- Learning Rotation-Equivariant Features for Visual Correspondence (CVPR 2023 · similarity 0.90)
- Variational Partial Group Convolutions for Input-Aware Partial Equivariance of Rotations and Color-Shifts (ICML 2024 · similarity 0.88)
- Relaxed Rotational Equivariance via G-Biases in Vision (AAAI 2025 · similarity 0.87)
- R2Det: Exploring Relaxed Rotation Equivariance in 2D Object Detection (ICLR 2025 · 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.