Research map / Large language models / Alignment and preferences
BeautyGRPO: Aesthetic Alignment for Face Retouching via Dynamic Path Guidance and Fine-Grained Preference Modeling
Jiachen Yang, Xianhui Lin, Yi Dong, Zebiao Zheng, Xing Liu, Hong Gu, Yanmei Fang
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 Dynamic Path Guidance to online RL (FlowGRPO) for retouching | FLUX.1 Kontext + LoRA · FFHQR | NIQE: 15.024 -> 10.831 | Improved |
| add Dynamic Path Guidance to online RL (FlowGRPO) for retouching | FLUX.1 Kontext + LoRA · FFHQR | MUSIQ: 4.271 -> 4.906 | Improved |
| add Dynamic Path Guidance to online RL (FlowGRPO) for retouching | FLUX.1 Kontext + LoRA · FFHQR | ArcFace: 0.882 -> 0.952 | Improved |
| apply BeautyGRPO RL on top of supervised LoRA fine-tuning | FLUX.1 Kontext + LoRA · FFHQR | NIMA: 4.694 -> 5.123 | Improved |
| apply BeautyGRPO RL on top of supervised LoRA fine-tuning | FLUX.1 Kontext + LoRA · FFHQR | FID: 1.735 -> 4.054 | Hurt |
| apply BeautyGRPO RL on top of supervised LoRA fine-tuning | FLUX.1 Kontext + LoRA · FFHQR | ArcFace: 0.973 -> 0.952 | Hurt |
| use the fine-grained retouching reward model instead of EditReward (ablation) | BeautyGRPO on FLUX.1 Kontext + LoRA · FFHQR | NIMA: 4.754 -> 5.123 | Improved |
| use the fine-grained retouching reward model instead of EditScore (ablation) | BeautyGRPO on FLUX.1 Kontext + LoRA · FFHQR | TOPIQ: 0.621 -> 0.676 | Improved |
| use the fine-grained retouching reward model instead of UnifiedReward-Edit (ablation) | BeautyGRPO on FLUX.1 Kontext + LoRA · FFHQR | MUSIQ: 4.571 -> 4.906 | Improved |
| apply BeautyGRPO to Qwen-Image-Edit on top of LoRA | Qwen-Image-Edit · FFHQR | MUSIQ: 4.745 -> 5.133 | Improved |
| apply BeautyGRPO to Qwen-Image-Edit on top of LoRA | Qwen-Image-Edit · FFHQR | TOPIQ: 0.595 -> 0.664 | Improved |
| increase DPG sampling segments K from 3 to 5 (ablation) | BeautyGRPO · FFHQR | MUSIQ / NIMA | No clear effect |
Closest papers by meaning
- RetouchGPT: LLM-based Interactive High-Fidelity Face Retouching via Imperfection Prompting (AAAI 2025 · similarity 0.87)
- PrefPaint: Aligning Image Inpainting Diffusion Model with Human Preference (NeurIPS 2024 · similarity 0.87)
- RetouchFormer: Semi-supervised High-Quality Face Retouching Transformer with Prior-Based Selective Self-Attention (AAAI 2024 · similarity 0.86)
- Aligning Vision Models with Human Aesthetics in Retrieval: Benchmarks and Algorithms (NeurIPS 2024 · similarity 0.86)
- Aesthetic Post-Training Diffusion Models from Generic Preferences with Step-by-step Preference Optimization (CVPR 2025 · similarity 0.86)
- Unlocking the Essence of Beauty: Advanced Aesthetic Reasoning with Relative-Absolute Policy Optimization (ICLR 2026 · similarity 0.85)
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.
