Research map / Label-efficient and robust learning / Recognition and long-tail
Permuted AdaIN: Reducing the Bias Towards Global Statistics in Image Classification
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 |
|---|---|---|---|
| Add pAdaIN during training | Image classifiers · CIFAR100 and ImageNet | Top-1 accuracy: 76.13 -> 77.82; 77.1 -> 77.7 | Helped |
| Add pAdaIN to source-only training | DeepLabv2 with ResNet101 · GTA5 → Cityscapes | mIoU: 37.0 -> 38.7 | Helped |
| Add pAdaIN to FADA | DeepLabv2 with ResNet101 · GTA5 → Cityscapes | mIoU: 49.2 -> 51.5 | Helped |
| Train with pAdaIN | ResNet18 and ResNet50 · PACS average accuracy | accuracy: 79.72 -> 82.51; 83.29 -> 85.36 | Helped |
| Train with pAdaIN | ResNet50 · ImageNet-C | mCE: 76.7 -> 72.8 | Helped |
| Add pAdaIN to Augmix | DenseNet-BC and ResNext-29 · CIFAR-100-C | mCE: 38.9 -> 37.5; 34.4 -> 31.6 | Helped |
| Increase p to 0.01 (ablation) | pAdaIN-trained ResNet18 and ResNet50 · CIFAR100 and ImageNet | accuracy | Helped |
| Apply pAdaIN only to deeper blocks rather than early blocks (ablation) | ResNet18 · CIFAR100 | accuracy: 76.1 -> 77.5 | Helped |
| Replace partner-image statistics with random normal statistics (ablation) | pAdaIN · CIFAR100 | accuracy: 57.3 | Hurt |
| Stop backpropagation through the current sample's statistics (ablation) | pAdaIN · CIFAR100 | accuracy: 77.8 -> 75.2 | Hurt |
| Backpropagate through partner-sample statistics (ablation) | pAdaIN · CIFAR100 | accuracy: 77.8 -> 77.6 | Mixed |
| Use a fixed permutation across all layers instead of drawing permutations independently (ablation) | pAdaIN · CIFAR100 | accuracy: 77.82 -> 68.02 | Hurt |
Closest papers by meaning
- Adaptive Convolutions for Structure-Aware Style Transfer (CVPR 2021 · similarity 0.85)
- ACNe: Attentive Context Normalization for Robust Permutation-Equivariant Learning (CVPR 2020 · similarity 0.85)
- iFAN: Image-Instance Full Alignment Networks for Adaptive Object Detection (AAAI 2020 · similarity 0.85)
- U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation (ICLR 2020 · similarity 0.85)
- CAT: Exploiting Inter-Class Dynamics for Domain Adaptive Object Detection (CVPR 2024 · similarity 0.84)
- Leveraging Normalization Layer in Adapters with Progressive Learning and Adaptive Distillation for Cross-Domain Few-Shot Learning (AAAI 2024 · 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.