Research map / Architectures and efficiency / CNN design
Deep Pyramidal Residual Networks
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 |
|---|---|---|---|
| gradually increase channels across residual units rather than sharply at downsampling | PyramidNet · CIFAR-10 | test error: 6.43 -> 4.58 | Helped |
| gradually increase channels across residual units rather than sharply at downsampling | PyramidNet · CIFAR-100 | test error: 25.16 -> 23.12 | Helped |
| increase widening factor from α=48 to α=84 (ablation) | PyramidNet · CIFAR-10 | test error: 4.58±0.06 -> 4.26±0.23 | Helped |
| increase widening factor from α=48 to α=84 (ablation) | PyramidNet · CIFAR-100 | test error: 23.12±0.04 -> 20.66±0.40 | Helped |
| increase widening factor across the reported PyramidNet settings (ablation) | PyramidNet · CIFAR-10 and CIFAR-100 | test error: 4.26±0.23 -> 3.73±0.04; 20.66±0.40 -> 18.25±0.10 | Helped |
| use bottleneck residual units (ablation) | PyramidNet · CIFAR-10 and CIFAR-100 | test error: 3.73±0.04 -> 3.48±0.20; 18.25±0.10 -> 17.01±0.39 | Helped |
| use the additive channel-growth schedule instead of multiplicative growth (ablation) | PyramidNet · CIFAR-10 and CIFAR-100 | test error | Mixed |
| use zero-padded identity-mapping shortcuts instead of projection-based shortcut variants (ablation) | PyramidNet · CIFAR-10 and CIFAR-100 | top-1 error: 5.03 -> 4.70; 23.48 -> 22.77 | Helped |
| remove the first ReLU from the pre-activation residual block (ablation) | ResNet · CIFAR-10 and CIFAR-100 | top-1 error: 5.82 -> 5.31; 25.06 -> 24.55 | Helped |
| add batch normalization after the final convolution (ablation) | ResNet · CIFAR-10 and CIFAR-100 | top-1 error: 5.82 -> 5.74; 25.06 -> 24.54 | Mixed |
| remove the first ReLU and add batch normalization after the final convolution (ablation) | ResNet · CIFAR-10 and CIFAR-100 | top-1 error: 5.82 -> 5.29; 25.06 -> 23.74 | Helped |
| remove the first ReLU from the pre-activation residual block (ablation) | PyramidNet · CIFAR-10 and CIFAR-100 | top-1 error: 5.15 -> 4.81; 24.40 -> 23.43 | Helped |
Closest papers by meaning
- Deep Residual Learning for Image Recognition (CVPR 2016 · similarity 0.89)
- Dilated Residual Networks (CVPR 2017 · similarity 0.89)
- The Unreasonable Effectiveness of Fully-Connected Layers for Low-Data Regimes (NeurIPS 2022 · similarity 0.88)
- Refining Architectures of Deep Convolutional Neural Networks (CVPR 2016 · similarity 0.88)
- Feature Pyramid Networks for Object Detection (CVPR 2017 · similarity 0.88)
- Spatially Adaptive Computation Time for Residual Networks (CVPR 2017 · 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.