Research map / Image generation / Editing and style transfer
Arbitrary Style Transfer in Real-Time With Adaptive Instance Normalization
Xun Huang, Serge Belongie
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 concatenation with the AdaIN layer to fuse content and style (ablation) | Enc-Concat-Dec · MS-COCO content / WikiArt style | content and style loss | Improved |
| remove BN layers from the decoder (ablation) | Enc-AdaIN-BNDec · MS-COCO content / WikiArt style | style and content loss | Improved |
| remove IN layers from the decoder (ablation) | Enc-AdaIN-INDec · MS-COCO content / WikiArt style | style and content loss | Improved |
| replace BN with IN in feed-forward style transfer training | single-style texture network · Style transfer training convergence | convergence speed | Improved |
| equalize luminance histogram so all training images share contrast | IN texture network · Style transfer training convergence | convergence speed | No clear effect |
| normalize all training images to the same style | IN vs BN texture network · Style transfer training convergence | IN advantage over BN | Hurt |
| use Gram matrix loss instead of IN-statistics style loss | AdaIN network · Style transfer | visual quality | No clear effect |
| use AdaIN output t as content target instead of content image features | AdaIN network · Style transfer training | convergence speed | Improved |
| use AdaIN network (arbitrary styles) instead of Chen and Schmidt style swap | AdaIN · Speed 512x512 | processing time (s): 3.214 -> 0.065 | Improved |
| use AdaIN network instead of Gatys optimization | AdaIN · Speed 512x512 | processing time (s): 46.75 -> 0.065 | Improved |
| compare AdaIN with single-style Ulyanov et al. on style/content loss | AdaIN · Style transfer loss (10 styles, 50 contents) | content and style loss | Mixed |
Closest papers by meaning
- Adaptive Convolutions for Structure-Aware Style Transfer (CVPR 2021 · similarity 0.93)
- Dynamic Instance Normalization for Arbitrary Style Transfer (AAAI 2020 · similarity 0.92)
- AdaAttN: Revisit Attention Mechanism in Arbitrary Neural Style Transfer (ICCV 2021 · similarity 0.91)
- Adjustable Real-time Style Transfer (ICLR 2020 · similarity 0.90)
- StyleStudio: Text-Driven Style Transfer with Selective Control of Style Elements (CVPR 2025 · similarity 0.88)
- Style Injection in Diffusion: A Training-free Approach for Adapting Large-scale Diffusion Models for Style Transfer (CVPR 2024 · 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.
