Research map / Label-efficient and robust learning / OOD and anomaly detection
CrossNorm and SelfNorm for Generalization Under Distribution Shifts
Zhiqiang Tang, Yunhe Gao, Yi Zhu, Zhi Zhang, Mu Li, Dimitris N. Metaxas
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 CrossNorm and SelfNorm together | WideResNet-40-2 etc. · CIFAR-100-C (mean over 4 backbones) | mCE (%): 55.6 -> 43.5 | Improved |
| add CrossNorm and SelfNorm on top of AugMix | AugMix · CIFAR-10-C (mean over 4 backbones) | mCE (%): 12.5 -> 10.3 | Improved |
| add CNSN and AugMix (with IBN) on ImageNet | ResNet50 AugMix · ImageNet-C | mCE (%): 68.4 -> 62.8 | Improved |
| add CrossNorm to FixMatch with RandAugment, 250 labels | FixMatch WideResNet-28-2 · CIFAR-10-C semi-supervised (250 labels, RandAugment) | mCE (%): 23.4 -> 16.8 | Improved |
| add CrossNorm to FixMatch with weak augmentation, 250 labels | FixMatch WideResNet-28-2 · CIFAR-10 semi-supervised (250 labels, weak aug) | mCE (%): 62.9 -> 56.5 | Improved |
| add CrossNorm and SelfNorm to FCN for domain generalization | FCN ResNet50 · GTA5->Cityscapes (target) | mIoU: 21.4 -> 36.5 | Improved |
| add CrossNorm to FCN, measured on source domain versus domain randomization | FCN ResNet50 · GTA5->Cityscapes (source GTA5) | mIoU: 49.0 -> 61.2 | Improved |
| add CrossNorm and SelfNorm to ConvNet sentiment classifier | GloVe ConvNet · IMDb->SST-2 (target SST-2) | accuracy (%): 71.9 -> 74.9 | Improved |
| add CrossNorm to ResNet50 for unsupervised domain adaptation | ResNet50 · Office-31 domain adaptation (avg) | accuracy (%): 76.1 -> 79.2 | Improved |
| use 2-instance instead of 1-instance CrossNorm (ablation) | CN WideResNet-40-2 · CIFAR-100-C (WideResNet-40-2) | mCE (%): 50.5 -> 48.8 | Improved |
| add style cropping to 2-instance CrossNorm (ablation) | CN WideResNet-40-2 · CIFAR-100-C (WideResNet-40-2) | mCE (%): 48.8 -> 47.9 | Improved |
| use SelfNorm instead of SE attention (ablation) | WideResNet-40-2 · CIFAR-100-C (WideResNet-40-2) | mCE (%): 52.3 -> 47.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AugMix — AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty (ICLR 2020) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Understanding and Improving Layer Normalization (NeurIPS 2019 · similarity 0.86)
- MetaNorm: Learning to Normalize Few-Shot Batches Across Domains (ICLR 2021 · similarity 0.85)
- A Fine-Grained Analysis on Distribution Shift (ICLR 2022 · similarity 0.85)
- SeeDNorm: Self-Rescaled Dynamic Normalization (ICLR 2026 · similarity 0.85)
- Beyond BatchNorm: Towards a Unified Understanding of Normalization in Deep Learning (NeurIPS 2021 · similarity 0.84)
- How Does Batch Normalization Help Optimization? (NeurIPS 2018 · 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.
