Research map / Image generation / Editing and style transfer
All-to-Key Attention for Arbitrary Style Transfer
Mingrui Zhu, Xiao He, Nannan Wang, Xiaoyu Wang, Xinbo Gao
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 all-to-all attention with all-to-key attention (ablation) | StyA2K · image style transfer | Content Loss: 0.73 -> 0.55 | Improved |
| add distributed attention (ablation) | StyA2K · image style transfer | Style Loss: 1.48 -> 1.04 | Improved |
| add progressive attention (ablation) | StyA2K · image style transfer | LPIPS: 0.59 -> 0.52 | Improved |
| add the coarse region-matching first step of progressive attention (ablation) | StyA2K · image style transfer | Content Loss: content 0.65 -> 0.55; style 0.73 -> 1.04 | Mixed |
| use multiple attention heads instead of one (ablation) | StyA2K · image style transfer | Content Loss: 0.65 -> 0.55 | Improved |
| increase lambda2 weight on attention-based feature matching loss (ablation) | StyA2K · image style transfer | Content Loss vs Style Loss: content 0.72 -> 0.49; style 0.75 -> 1.38 (lambda2 0.5 -> 2) | Mixed |
| StyA2K vs all-to-all methods in style loss | StyA2K · image style transfer | Style Loss: SANet 0.99 vs StyA2K 1.04 | Mixed |
| apply A2K to video frames without temporal training | StyA2K · video style transfer (5 clips) | optical flow error: 4.94 -> 4.24 | Improved |
| use sparse blocked attention instead of dense all-to-all attention | StyA2K · 512x512 inference | inference time (sec/image): AdaAttN 0.0269 -> 0.0143 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- AdaAttN — AdaAttN: Revisit Attention Mechanism in Arbitrary Neural Style Transfer (ICCV 2021) · better on 5 of 5 shared results
- StyTr2 — StyTr2: Image Style Transfer With Transformers (CVPR 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SCSA: A Plug-and-Play Semantic Continuous-Sparse Attention for Arbitrary Semantic Style Transfer (CVPR 2025 · similarity 0.91)
- HSI: A Holistic Style Injector for Arbitrary Style Transfer (CVPR 2025 · similarity 0.89)
- Style Injection in Diffusion: A Training-free Approach for Adapting Large-scale Diffusion Models for Style Transfer (CVPR 2024 · similarity 0.88)
- Style Aligned Image Generation via Shared Attention (CVPR 2024 · similarity 0.87)
- Attention Distillation: A Unified Approach to Visual Characteristics Transfer (CVPR 2025 · similarity 0.86)
- StyleStudio: Text-Driven Style Transfer with Selective Control of Style Elements (CVPR 2025 · similarity 0.86)
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.
