Research map / Architectures and efficiency / CNN design
Deep Diffeomorphic Transformer Networks
Nicki Skafte Detlefsen, Oren Freifeld, Søren Hauberg
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 |
|---|---|---|---|
| Add an affine and CPAB transformation layer | CNN · Distorted MNIST RTS | accuracy: 0.992 -> 0.996 | Improved |
| Add an affine and CPAB transformation layer | CNN · Distorted MNIST CPAB | accuracy: 0.989 -> 0.993 | Improved |
| Add a spatial transformer | CNN · MNIST | classification performance: 0.991 -> 0.996 | Improved |
| Add a CPAB transformer layer | CNN · Fashion MNIST | classification performance: 0.922 -> 0.917 | Hurt |
| Add affine followed by CPAB transformations | CNN · Fashion MNIST | classification performance: 0.922 -> 0.913 | Hurt |
| Add a CPAB transformer layer | CNN · CIFAR10 | classification performance: 0.870 -> 0.895 | Improved |
| Add a CPAB transformer layer | CNN · CIFAR100 | classification performance: 0.642 -> 0.659 | Improved |
| Add affine followed by CPAB transformations | Siamese network · LFW restricted | classification performance: 0.788 -> 0.893 | Improved |
| Add affine followed by CPAB transformations | CNN · CelebA | classification performance: 0.712 -> 0.772 | Improved |
| Add affine followed by CPAB transformations | CNN · LFW unrestricted | test accuracy: 0.8930 ± 0.0028 -> 0.9543 ± 0.0046 | Improved |
| Replace non-diffeomorphic affine transformations with diffeomorphic affine transformations (ablation) | Affine spatial transformer · LFW restricted | classification performance: 0.840 -> 0.842 | No clear effect |
| Replace ST-CPAB with ST-Affine+CPAB (ablation) | CNN · LFW restricted | classification performance: 0.878 -> 0.893 | Improved |
Closest papers by meaning
- Recursive Spatial Transformer (ReST) for Alignment-Free Face Recognition (ICCV 2017 · similarity 0.87)
- Diffeomorphic Learning (JMLR 2020 · similarity 0.87)
- Fast Symmetric Diffeomorphic Image Registration with Convolutional Neural Networks (CVPR 2020 · similarity 0.86)
- Neural Convolutional Surfaces (CVPR 2022 · similarity 0.86)
- Building Transformation Layers for Riemannian Neural Networks (NeurIPS 2026 · similarity 0.85)
- Neuroexplicit Diffusion Models for Inpainting of Optical Flow Fields (ICML 2024 · similarity 0.85)
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.
