Research map / Architectures and efficiency / CNN design
SRM: A Style-Based Recalibration Module for Convolutional Neural Networks
HyunJae Lee, Hyo-Eun Kim, Hyeonseob Nam
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 StdPool to AvgPool in style pooling (ablation) | SRM · ImageNet-1K val (ResNet-50) | top-1 accuracy (%): 76.58 -> 77.13 | Improved |
| use StdPool + AvgPool instead of AvgPool + MaxPool (ablation) | SRM · ImageNet-1K val (ResNet-50) | top-1 accuracy (%): 76.35 -> 77.13 | Improved |
| add MaxPool as the only pooling (ablation) | SRM · ImageNet-1K val (ResNet-50) | top-1 accuracy (%): 75.89 -> 75.87 | No clear effect |
| use channel-wise FC (CFC) instead of MLP for style integration (ablation) | SRM · ImageNet-1K val (ResNet-50) | top-1 accuracy (%): 76.75 -> 76.91 | Improved |
| add batch normalization after CFC (ablation) | SRM · ImageNet-1K val (ResNet-50) | top-1 accuracy (%): 76.91 -> 77.13 | Improved |
| add batch normalization after MLP (ablation) | SRM · ImageNet-1K val (ResNet-50) | top-1 accuracy (%): 76.75 -> 76.68 | Hurt |
| add SRM to ResNet-50 instead of SE (ablation) | ResNet-50 · ImageNet-1K val | top-1 accuracy (%): 76.80 -> 77.13 | Improved |
| add SRM to ResNet-101 | ResNet-101 · ImageNet-1K val | top-1 accuracy (%): 77.40 -> 78.47 | Improved |
| add GE to ResNet-101 | ResNet-101 · ImageNet-1K val | top-1 accuracy (%): 77.40 -> 77.36 | No clear effect |
| add SRM to ResNet-56 | ResNet-56 · CIFAR-100 | top-1 accuracy (%): 74.76 -> 76.93 | Improved |
| add SRM to ResNet-50 trained on Stylized-ImageNet | ResNet-50 · Stylized-ImageNet | top-1 accuracy (%): 53.93 -> 60.69 | Improved |
| add SRM to ResNet-18 on Office-Home | ResNet-18 · Office-Home | accuracy (%): 55.47 -> 59.47 | Improved |
Closest papers by meaning
- StyleBank: An Explicit Representation for Neural Image Style Transfer (CVPR 2017 · similarity 0.87)
- Rethinking and Improving the Robustness of Image Style Transfer (CVPR 2021 · similarity 0.87)
- Explicit Filterbank Learning for Neural Image Style Transfer and Image Processing (TPAMI 2020 · similarity 0.86)
- SaMam: Style-aware State Space Model for Arbitrary Image Style Transfer (CVPR 2025 · similarity 0.85)
- StyleSpace Analysis: Disentangled Controls for StyleGAN Image Generation (CVPR 2021 · similarity 0.85)
- StyTr2: Image Style Transfer With Transformers (CVPR 2022 · 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.
