Research map / Image generation / Text-to-image diffusion
Style-GRPO: Semantic-Aware Preference Optimization for Image Style Transfer Guided by Reward Modeling
Jianbin Zhao, Chaoran Feng, Miao Yu, Yingtao Li, Zhenyu Tang, Wangbo Yu, Yian Zhao, Xiaomin Li, Li Yuan, Yonghong Tian
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 SFT on StyleReward-Dataset to FLUX.1 Kontext (ablation) | FLUX.1 Kontext · ImgEdit (GPT-4o) | score: 4.55 -> 4.67 | Improved |
| add GRPO post-training with StyleScore directly to FLUX.1 Kontext (no SFT) (ablation) | FLUX.1 Kontext · ImgEdit (GPT-4o) | score: 4.55 -> 4.68 | Improved |
| add GRPO post-training after SFT (ablation) | SFT model · ImgEdit (GPT-4o) | score: 4.67 -> 4.74 | Improved |
| add GRPO post-training after SFT (ablation) | SFT model · ImgEdit (Gemini-2.5-Pro) | score: 4.34 -> 4.46 | Improved |
| add SFT before GRPO instead of GRPO only (ablation) | Style-GRPO · StyleReward-Dataset benchmark (StyleScore) | StyleScore: 3.85 -> 3.91 | Improved |
| add GRPO only (no SFT) versus SFT only on Gemini judge (ablation) | FLUX.1 Kontext · ImgEdit (Gemini-2.5-Pro) | score: 4.29 -> 4.30 | No clear effect |
| train StyleScore reward model on StyleReward-Dataset instead of using general VLM | Qwen2.5-VL-7B reward model · 500 test pairs | preference accuracy: 65.2% -> 98.6% | Improved |
| apply Style-GRPO to FLUX.1 Kontext | FLUX.1 Kontext · AnyEdit global style transfer | L1 distance: 0.2457 -> 0.0944 | Improved |
| apply Style-GRPO to FLUX.1 Kontext | FLUX.1 Kontext · AnyEdit global style transfer | CLIPtext: 0.1857 -> 0.1664 | Hurt |
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.
- InstructPix2Pix — InstructPix2Pix: Learning To Follow Image Editing Instructions (CVPR 2023) · better on 2 of 2 shared results
- ImageReward — ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation (NeurIPS 2023) · better on 1 of 1 shared result
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
- SpatialFlow-GRPO: Where Spatial Credit Drives Image Editing (NeurIPS 2026 · similarity 0.87)
- GoT-R1: Unleashing Reasoning Capability of Autoregressive Visual Generation with Reinforcement Learning (ICLR 2026 · similarity 0.87)
- Learning to Sample Effective and Diverse Prompts for Text-to-Image Generation (CVPR 2025 · similarity 0.87)
- Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation (NeurIPS 2025 · similarity 0.87)
- SpatialReward: Bridging the Perception Gap in Online RL for Image Editing via Explicit Spatial Reasoning (ICML 2026 · similarity 0.87)
- SSMG: Spatial-Semantic Map Guided Diffusion Model for Free-Form Layout-to-Image Generation (AAAI 2024 · 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.
