Research map / Label-efficient and robust learning / Recognition and long-tail
Batch-Instance Normalization for Adaptively Style-Invariant Neural Networks
Hyeonseob Nam, Hyo-Eun Kim
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 with learnable BIN | ResNet-110 · CIFAR-10 test | top-1 accuracy (%): 93.72 ± 0.18 -> 94.29 ± 0.09 | Improved |
| Replace BN with learnable BIN | ResNet-110 · CIFAR-100 test | top-1 accuracy (%): 74.26 ± 0.33 -> 75.88 ± 0.30 | Improved |
| Replace BN with learnable BIN | ResNet-18 · ImageNet validation | top-1 accuracy (%): 69.89 -> 70.68 | Improved |
| Replace BN with learnable BIN | ResNet-18 · Office-Home mixed-domain classification | average accuracy (%): 78.95 -> 80.08 | Improved |
| Replace BN with learnable BIN | DANN with ResNet-18 · Office-Home domain adaptation, 12 transfer tasks | accuracy (%) | Mixed |
| Replace BN with IN (ablation) | ResNet-110 · CIFAR-100 | top-1 accuracy | Hurt |
| Use a fixed 0.5 BN/IN mixture instead of a learned per-channel gate (ablation) | ResNet-110 · CIFAR-100 | top-1 accuracy | Hurt |
| Use BIN instead of IN (ablation) | Style transfer network · MS-COCO content images and artistic styles | content loss | Improved |
| Use BIN instead of IN (ablation) | Style transfer network · MS-COCO content images and artistic styles | style loss | Hurt |
| Use BIN instead of BN+IN (ablation) | Style transfer network · MS-COCO content images and artistic styles | style loss | Improved |
Closest papers by meaning
- Interpolating Convolutional Neural Networks Using Batch Normalization (ECCV 2018 · similarity 0.88)
- Dynamic Instance Normalization for Arbitrary Style Transfer (AAAI 2020 · similarity 0.88)
- Adaptive Convolutions for Structure-Aware Style Transfer (CVPR 2021 · similarity 0.86)
- AdaAttN: Revisit Attention Mechanism in Arbitrary Neural Style Transfer (ICCV 2021 · similarity 0.86)
- Arbitrary Style Transfer in Real-Time With Adaptive Instance Normalization (ICCV 2017 · similarity 0.86)
- High-Performance Large-Scale Image Recognition Without Normalization (ICML 2021 · 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.
