Research map / Architectures and efficiency / CNN design
Log-Polar Space Convolution Layers
Bing Su, Ji-Rong Wen
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 |
|---|---|---|---|
| replace first conv layers with LPSC | AlexNet · CIFAR-10 | top-1 accuracy: 77.43 -> 78.44 | Improved |
| replace first conv layers with LPSC | AlexNet · CIFAR-100 | top-1 accuracy: 43.98 -> 47.43 | Improved |
| replace first conv layers with dilated conv | AlexNet · CIFAR-10 | top-1 accuracy: 77.43 -> 75.42 | Hurt |
| replace conv with LPSC instead of dilated conv | AlexNet · CIFAR-100 | top-1 accuracy: 44.43 -> 47.43 | Improved |
| replace added first conv with LPSC | VGGNet-19 · CIFAR-10 | top-1 accuracy: 93.54 -> 93.92 | Improved |
| replace added first conv with LPSC | VGGNet-19 · CIFAR-100 | top-1 accuracy: 72.41 -> 73.13 | Improved |
| replace first conv with LPSC | ResNet-20 · CIFAR-10 | top-1 accuracy: 91.66 -> 91.81 | Improved |
| use cross convolution (LPSC-CC, p=50) instead of plain LPSC (ablation) | ResNet-20 · CIFAR-100 | top-1 accuracy: 67.63 -> 68.09 | Improved |
| use dilated conv instead of conventional conv | ResNet-20 · CIFAR-100 | top-1 accuracy: 67.56 -> 66.90 | Hurt |
| replace first conv with LPSC | ResNet-110 · CIFAR-100 | top-1 accuracy: 73.50 -> 73.54 | No clear effect |
| use LPSC-CC (p=50, first three layers) on ResNet-110 | ResNet-110 · CIFAR-100 | top-1 accuracy: 73.50 -> 74.13 | Improved |
| LPSC implemented via standard conv modules | ResNet-110 · CIFAR-100 | inference time: 0.025 -> 0.036 | Hurt |
Closest papers by meaning
- Non-Linear Convolution Filters for CNN-Based Learning (ICCV 2017 · similarity 0.86)
- Convolution of Convolution: Let Kernels Spatially Collaborate (CVPR 2022 · similarity 0.85)
- KernelDNA: Cross-Layer Kernel Sharing via Decoupled Neural Adapters (NeurIPS 2026 · similarity 0.85)
- A General Theory of Equivariant CNNs on Homogeneous Spaces (NeurIPS 2019 · similarity 0.85)
- Brain-inspired $L_p$-Convolution benefits large kernels and aligns better with visual cortex (ICLR 2025 · similarity 0.84)
- LSNet: See Large, Focus Small (CVPR 2025 · similarity 0.83)
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.
