Research map / Architectures and efficiency / CNN design
Rotation Equivariant Vector Field Networks
Diego Marcos, Michele Volpi, Nikos Komodakis, Devis Tuia
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 vector field feature maps instead of scalar field (orientation pooling keeping only magnitude) (ablation) | RotEqNet · MNIST-rot | error rate (%): 2.01 -> 1.09 | Improved |
| add test-time rotation augmentation | RotEqNet · MNIST-rot | error rate (%): 1.09 -> 1.01 | Improved |
| add training data augmentation | standard CNN trained on MNIST · MNIST-rot | error rate (%): 57% -> 2.3% | Improved |
| add training data augmentation | RotEqNet trained on MNIST · MNIST-rot | error rate (%): 20% -> 1.1% | Improved |
| add training data augmentation | standard CNN trained on MNIST-rot · MNIST-rot | error rate (%): 4.9% -> 2.2% | Improved |
| add training data augmentation | RotEqNet trained on MNIST-rot · MNIST-rot | error rate (%): 1.4% -> 1.1% | Mixed |
| replace standard CNN (10x parameters) with RotEqNet | MNIST-rot training without augmentation · MNIST-rot | error rate (%): 4.9% -> 1.4% | Improved |
| add training data augmentation | standard CNN · ISBI 2012 (validation) | score: 0.9232 -> 0.9572 | Improved |
| add training data augmentation | RotEqNet · ISBI 2012 (validation) | score: 0.9726 -> 0.9790 | Improved |
| ensemble three models (N=1,2,3) instead of single N=2 model | RotEqNet · ISBI 2012 test (Rand. Thin) | Rand score (thinned): 0.9599 -> 0.9712 | Improved |
| ensemble three models (N=1,2,3) instead of single N=2 model | RotEqNet · ISBI 2012 test (Inf. Thin) | information score (thinned): 0.9806 -> 0.9865 | Improved |
| increase number of orientations R beyond 17 at test time (ablation) | RotEqNet · Car orientation estimation | average error (degrees) | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TI-pooling — TI-Pooling: Transformation-Invariant Pooling for Feature Learning in Convolutional Neural Networks (CVPR 2016) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Harmonic Networks: Deep Translation and Rotation Equivariance (CVPR 2017 · similarity 0.90)
- Empowering Networks With Scale and Rotation Equivariance Using A Similarity Convolution (ICLR 2023 · similarity 0.90)
- RRL: Regional Rotate Layer in Convolutional Neural Networks (AAAI 2022 · similarity 0.90)
- Rotation-Equivariant Self-Supervised Method in Image Denoising (CVPR 2025 · similarity 0.88)
- Relaxed Rotational Equivariance via G-Biases in Vision (AAAI 2025 · similarity 0.88)
- A Probabilistic Approach to Learning the Degree of Equivariance in Steerable CNNs (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.
