Research map / Image generation / Editing and style transfer
Mixture of Style Experts for Diverse Image Stylization
Shihao Zhu, Ziheng Ouyang, Yijia Kang, Qilong Wang, Mi Zhou, Bo Li, Ming-Ming Cheng, Qibin Hou
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 |
|---|---|---|---|
| add pre-trained style encoder to MoE router (ablation) | StyleExpert · Style test set (21 unseen styles) | Qwen Semantic: 71.43 -> 75.12 | Improved |
| add pre-trained style encoder to MoE router (ablation) | StyleExpert · Style test set (21 unseen styles) | CSD: 66.70 -> 73.18 | Improved |
| add pre-trained style encoder to MoE router (ablation) | StyleExpert · Style test set (21 unseen styles) | DreamSim: 38.54 -> 28.18 | Improved |
| replace LoRA-only fine-tuning with encoder-guided MoE (ablation) | StyleExpert · Style test set (21 unseen styles) | CSD: 70.88 -> 73.18 | Improved |
| MoE without style encoder vs LoRA-only (ablation) | StyleExpert · Style test set (21 unseen styles) | DreamSim: 36.77 -> 38.54 | Hurt |
| add style encoder to MoE training (ablation) | StyleExpert · Style test set (21 unseen styles) | convergence | Improved |
| StyleExpert vs USO on content structure | StyleExpert vs USO · Style test set (21 unseen styles) | DINO: 84.03 -> 64.72 | Hurt |
| StyleExpert vs strongest baselines on semantic stylization | StyleExpert · Style test set (21 unseen styles) | Qwen Semantic: 42.74 -> 75.12 | Improved |
| style-encoder latents guide router expert choice | StyleExpert · Style test set (21 unseen styles) | expert IoU: 10.60% -> 33.18% | 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.
- OmniStyle — OmniStyle: Filtering High Quality Style Transfer Data at Scale (CVPR 2025) · better on 4 of 5 shared results
- CSGO — CSGO: Content-Style Composition in Text-to-Image Generation (NeurIPS 2025) · better on 4 of 5 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
- StyleFormer: Real-Time Arbitrary Style Transfer via Parametric Style Composition (ICCV 2021 · similarity 0.88)
- Learning to Stylize by Learning to Destylize: A Scalable Paradigm for Supervised Style Transfer (ECCV 2026 · similarity 0.88)
- Multi-StyleGS: Stylized Gaussian Splatting with Multiple Styles (AAAI 2025 · similarity 0.87)
- OmniStyle: Filtering High Quality Style Transfer Data at Scale (CVPR 2025 · similarity 0.87)
- SigStyle: Signature Style Transfer via Personalized Text-to-Image Models (AAAI 2025 · similarity 0.86)
- Morpheus: Text-Driven 3D Gaussian Splat Shape and Color Stylization (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.
