Research map / Architectures and efficiency / CNN design
On Translation Invariance in CNNs: Convolutional Layers Can Exploit Absolute Spatial Location
Check 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 F-Conv instead of S-Conv | ResNet-50 on ImageNet with 50 images per class · ImageNet, 50 images per class | accuracy: 26.4% -> 31.1% | Helped |
| Use F-Conv instead of S-Conv | HardNet · Patch matching, 50k triplet patches | FPR at 0.95 true positive recall: 0.145 -> 0.083 | Helped |
| Use F-Conv with zero padding instead of S-Conv with zero padding | 4-layer fully convolutional network · Red-Green Dissimilar test | accuracy: 8.4 ± 0.7 -> 89.7 ± 0.5 | Helped |
| Use F-Conv with zero padding instead of S-Conv with circular padding | 4-layer fully convolutional network · Red-Green Dissimilar test | accuracy: 73.7 ± 1.0 -> 89.7 ± 0.5 | Helped |
| Use F-Conv with BlurPool Tri-3 instead of S-Conv | ResNet-34 · ImageNet subset, diagonal shift | accuracy: 82.06 -> 86.91 | Helped |
| Use F-Conv with BlurPool Tri-3 instead of S-Conv | ResNet-34 · ImageNet subset, consistency | consistency: 87.62 -> 91.53 | Helped |
| Use F-Conv instead of S-Conv | 3D ResNet-18 · UCF101 | accuracy: 38.6 -> 40.6 | Helped |
| Use F-Conv instead of S-Conv | 3D ResNet-34 · UCF101 | accuracy: 37.0 -> 46.9 | Helped |
| Use F-Conv instead of S-Conv | 3D ResNet-50 · UCF101 | accuracy: 36.2 -> 44.1 | Helped |
| Use F-Conv instead of S-Conv | 3D ResNet-18 · HMDB51 | accuracy: 16.1 -> 19.3 | Helped |
| Use F-Conv instead of S-Conv | 3D ResNet-34 · HMDB51 | accuracy: 15.2 -> 18.3 | Helped |
| Use F-Conv instead of S-Conv | 3D ResNet-50 · HMDB51 | accuracy: 14.3 -> 19.0 | Helped |
Closest papers by meaning
- Warped Convolutions: Efficient Invariance to Spatial Transformations (ICML 2017 · similarity 0.88)
- Role of Locality and Weight Sharing in Image-Based Tasks: A Sample Complexity Separation between CNNs, LCNs, and FCNs (ICLR 2024 · similarity 0.86)
- Mechanisms of Object Localization in Vision-Language Models (CVPR 2026 · similarity 0.85)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.84)
- Where Am I and What Will I See: An Auto-Regressive Model for Spatial Localization and View Prediction (ICLR 2025 · similarity 0.84)
- Tilt your Head: Activating the Hidden Spatial-Invariance of Classifiers (ICML 2024 · similarity 0.84)
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.