Research map / Label-efficient and robust learning / Recognition and long-tail
Style-Aware Normalized Loss for Improving Arbitrary Style Transfer
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 |
|---|---|---|---|
| Train GoogleMagenta with style-aware normalized loss instead of classic loss | GoogleMagenta · PBN test / ImageNet test | deception rate: 16.84% -> 35.38%; 110% relative improvement | Helped |
| Train AdaIN with style-aware normalized loss instead of classic loss | AdaIN · PBN test / ImageNet test | deception rate: 10.78% -> 22.18%; 106% relative improvement | Helped |
| Train LinearTransfer with style-aware normalized loss instead of classic loss | LinearTransfer · PBN test / ImageNet test | deception rate: 18.58% -> 39.18%; 111% relative improvement | Helped |
| Train SANet with style-aware normalized loss instead of classic loss | SANet · PBN test / ImageNet test | deception rate: 18.08% -> 33.16%; 83% relative improvement | Helped |
| Train AST models with style-aware normalized loss instead of classic loss | GoogleMagenta, AdaIN, LinearTransfer, and SANet · Human evaluation | vote preference: 28.2% -> 55.9% | Helped |
| Normalize each style loss by its theoretical supremum during training | GoogleMagenta, AdaIN, LinearTransfer, and SANet · AST test samples | classic style loss: GoogleMagenta: 3.40 $\times 10^8$ -> 3.35 $\mathbf{\times 10^8}$; AdaIN: 6.62 $\times 10^8$ -> 6.58 $\mathbf{\times 10^8 | Helped |
| Use the classic style loss rather than style-aware normalization | AST style transfer · Human-rated stylization quality | correlation with human score: Average correlation: -0.167 (classic); 0.165 (balanced) | Hurt |
| Use style interpolation between the content and style features as an over-stylization remedy | GoogleMagenta · Qualitative comparison | content preservation with proper style transfer | Hurt |
Closest papers by meaning
- MetaStyle: Three-Way Trade-off among Speed, Flexibility, and Quality in Neural Style Transfer (AAAI 2019 · similarity 0.90)
- Arbitrary Style Transfer With Style-Attentional Networks (CVPR 2019 · similarity 0.90)
- AesFA: An Aesthetic Feature-Aware Arbitrary Neural Style Transfer (AAAI 2024 · similarity 0.90)
- OmniStyle: Filtering High Quality Style Transfer Data at Scale (CVPR 2025 · similarity 0.88)
- StyleStudio: Text-Driven Style Transfer with Selective Control of Style Elements (CVPR 2025 · similarity 0.87)
- SigStyle: Signature Style Transfer via Personalized Text-to-Image Models (AAAI 2025 · similarity 0.87)
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.