Research map / Image generation / Image restoration
Instance Enhancement Batch Normalization: An Adaptive Regulator of Batch Noise
Read the paperCheck 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 |
|---|---|---|---|
| Replace BN with IEBN | ResNet164 · CIFAR100 | top1-acc.: 74.29 -> 77.09 | Helped |
| Replace BN with IEBN | ResNet50 · ImageNet | top1-acc.: 76.01 -> 77.10 | Helped |
| Replace BN with IEBN | PreResNet164 · CIFAR100 | top1-acc.: 76.56 -> 77.27 | Mixed |
| Replace BN with IEBN | ResNext29,8x64 · CIFAR10 | top1-acc.: 96.11 -> 96.26 | Mixed |
| Replace BN with IEBN | ResNet164 · CIFAR100 with constant noise (0.8,0.8) | test accuracy: 45.42$_{(\pm31.42)}$ -> 75.42$_{(\pm0.08)}$ | Helped |
| Replace BN with IEBN | ResNet164 · CIFAR100 mixed with MNIST, C+3× M | test accuracy: 71.54 -> 76.03 | Helped |
| Use identity processing instead of trainable linear transformation for the squeezed feature (ablation) | IEBN on ResNet164 · CIFAR100 | test accuracy: 77.09($\pm0.15$) -> 67.53($\pm$2.49) | Hurt |
| Use a fully connected gate instead of the per-channel linear gate (ablation) | IEBN on ResNet164 · CIFAR100 | test accuracy: 77.09($\pm0.15$) -> 76.11($\pm0.28$) | Hurt |
| Apply the instance gate to beta only instead of gamma (ablation) | IEBN on ResNet164 · CIFAR100 | test accuracy: 77.09($\pm0.15$) -> 75.03($\pm0.54$) | Hurt |
| Apply the instance gate to both gamma and beta instead of gamma only (ablation) | IEBN on ResNet164 · CIFAR100 | test accuracy: 77.09($\pm0.15$) -> 77.02($\pm0.08$) | No clear effect |
| Initialize the gate parameters to (hat gamma, hat beta) = (0, -1) rather than (0, 0) (ablation) | IEBN on ResNet164 · CIFAR100 | test accuracy: 76.21 -> 77.09 | Helped |
| Initialize the gate parameters to (hat gamma, hat beta) = (1, 1) rather than (0, -1) (ablation) | IEBN on ResNet164 · CIFAR100 | test accuracy: 77.09 -> 68.5 | Hurt |
Closest papers by meaning
- Understanding Batch Normalization (NeurIPS 2018 · similarity 0.89)
- Ghost Noise for Regularizing Deep Neural Networks (AAAI 2024 · similarity 0.89)
- Filter Response Normalization Layer: Eliminating Batch Dependence in the Training of Deep Neural Networks (CVPR 2020 · similarity 0.88)
- Towards Understanding Regularization in Batch Normalization (ICLR 2019 · similarity 0.88)
- Batch normalization is sufficient for universal function approximation in CNNs (ICLR 2024 · similarity 0.88)
- Four Things Everyone Should Know to Improve Batch Normalization (ICLR 2020 · similarity 0.88)
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.