Research map / Image generation / Editing and style transfer
Learning to Stylize by Learning to Destylize: A Scalable Paradigm for Supervised Style Transfer
Ye Wang, Zili Yi, Yibo Zhang, Peng Zheng, Xuping Xie, Jiang Lin, Yijun Li, Yilin Wang, Rui Ma
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 |
|---|---|---|---|
| fine-tune on DeStyle-350K | Flux.2-Klein-4B · BCS-Bench | Qwen-L (leakage, lower better): 0.4380 -> 0.0025 | Improved |
| fine-tune on DeStyle-350K | Flux.2-Klein-4B · BCS-Bench | DINO: 0.4985 -> 0.6930 | Improved |
| fine-tune on DeStyle-350K | Flux.2-Klein-4B · BCS-Bench | CSD: 0.6362 -> 0.4999 | Hurt |
| fine-tune on DeStyle-350K | Flux.2-Klein-4B · BCS-Bench | Qwen-S: 3.5424 -> 3.1937 | Hurt |
| fine-tune on DeStyle-350K | Flux.2-Klein-9B · BCS-Bench | Qwen-C: 3.7121 -> 4.5610 | Improved |
| fine-tune on DeStyle-350K | Flux.2-Klein-9B · BCS-Bench | Qwen-L (leakage, lower better): 0.4748 -> 0.0059 | Improved |
| fine-tune on DeStyle-350K | Flux.2-Klein-9B · BCS-Bench | CSD: 0.6240 -> 0.5485 | Hurt |
| fine-tune on DeStyle-350K with LoRA | Qwen-Image-Edit-2511 · BCS-Bench | Qwen-C: 3.2562 -> 4.5013 | Improved |
| fine-tune on DeStyle-350K with LoRA | Qwen-Image-Edit-2511 · BCS-Bench | Qwen-S: 3.6436 -> 3.8954 | Mixed |
| train on authentic destylized triplets (DeStyle-350K) instead of synthetic OmniStyle150K pairs | dataset quality · Qwen3-VL-Plus dataset scoring | style consistency score: 3.78 -> 4.54 | Improved |
| use DeStyle-350K-tuned Flux.2-Klein-9B instead of OmniStyle | style transfer model · BCS-Bench | Qwen-S: 2.8840 -> 4.1140 | Improved |
| use DeStyle-350K-tuned Flux.2-Klein-9B instead of USO | style transfer model · BCS-Bench | Qwen-S: 2.8612 -> 4.1140 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OmniStyle — OmniStyle: Filtering High Quality Style Transfer Data at Scale (CVPR 2025) · better on 8 of 9 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
- OmniStyle: Filtering High Quality Style Transfer Data at Scale (CVPR 2025 · similarity 0.90)
- StyleShot: A Snapshot on Any Style (TPAMI 2025 · similarity 0.89)
- StylerDALLE: Language-Guided Style Transfer Using a Vector-Quantized Tokenizer of a Large-Scale Generative Model (ICCV 2023 · similarity 0.89)
- CSGO: Content-Style Composition in Text-to-Image Generation (NeurIPS 2025 · similarity 0.88)
- SigStyle: Signature Style Transfer via Personalized Text-to-Image Models (AAAI 2025 · similarity 0.88)
- Conditional Balance: Improving Multi-Conditioning Trade-Offs in Image Generation (CVPR 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.
