Research map / Detection and segmentation / CNN features
Switchable Whitening for Deep Representation Learning
Xingang Pan, Xiaohang Zhan, Jianping Shi, Xiaoou Tang, Ping Luo
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 BN layers with SW^a | ResNet50 · ImageNet val | top-1 error (%): 23.58 -> 22.10 | Improved |
| use SW^b instead of SN | ResNet50 · ImageNet val | top-1 error (%): 23.10 -> 22.07 | Improved |
| add instance whitening to batch whitening (SW^a vs BW) (ablation) | ResNet20 · CIFAR-10 | test error (%): 8.28 -> 7.64 | Improved |
| add BN, IN and LN statistics to SW^a (SW^b) (ablation) | ResNet50 · ImageNet val | top-1 error (%): 22.10 -> 22.07 | No clear effect |
| add BN, IN and LN statistics to SW^a (SW^b) (ablation) | ResNet20 · CIFAR-100 | test error (%): 31.00 -> 30.87 | No clear effect |
| use independent importance weights for mean and covariance instead of shared (ablation) | SW · CIFAR-10 ResNet20 | test error (%): 8.17 -> 7.64 | Improved |
| use switchable whitening instead of instance whitening only (ablation) | ResNet20 · CIFAR-10 | test error (%): 12.57 -> 7.64 | Improved |
| replace BN layers with SW^a | ResNet50 · ADE20K val (single-scale) | mIoU (%): 36.6 -> 39.8 | Improved |
| replace BN layers with SW^a | ResNet50 · Cityscapes val (single-scale) | mIoU (%): 72.1 -> 76.2 | Improved |
| replace BN with batch whitening | ResNet50 · ADE20K val (single-scale) | mIoU (%): 36.6 -> 35.9 | Hurt |
| add SW^a layers | PSPNet101 · ADE20K val | mIoU (%): 43.59 -> 45.33 | Improved |
| compute whitening with Newton iteration instead of SVD (ablation) | ResNet50-SW^a · ImageNet val | top-1 error (%): 22.10 -> 22.07 | No clear effect |
Closest papers by meaning
- Improving Generalization of Batch Whitening by Convolutional Unit Optimization (ICCV 2021 · similarity 0.90)
- Stochastic Whitening Batch Normalization (CVPR 2021 · similarity 0.88)
- An Investigation Into the Stochasticity of Batch Whitening (CVPR 2020 · similarity 0.87)
- Switchable Normalization for Learning-to-Normalize Deep Representation (TPAMI 2019 · similarity 0.87)
- Differentiable Learning-to-Normalize via Switchable Normalization (ICLR 2019 · similarity 0.87)
- Normalize Filters! Classical Wisdom for Deep Vision (NeurIPS 2025 · 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.
