Research map / Image generation / Text-to-image diffusion
Preference-Enhanced Reinforcement Learning for Pluralistic Image Inpainting
Peng Zhou, Muqi Huang, Tianshuo Qu, Jingyang Wang, kun Zhou, Chuan Li, Feng Shi, Shi Chen, Yun Xiong
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 |
|---|---|---|---|
| train Flux.1-fill with GRPO and preference reward instead of base | Flux.1-fill · FluxBench | ImageReward x10^2: -32.53 -> 31.41 | Improved |
| train Flux.1-fill with GRPO and preference reward instead of base | Flux.1-fill · EditBench | ImageReward x10^2: -33.36 -> 8.98 | Improved |
| train Flux.1-fill with GRPO and preference reward instead of base | Flux.1-fill · BrushBench | ImageReward x10^2: 48.64 -> 55.03 | Improved |
| train SD1.5-inpainting with GRPO instead of base | SD1.5-inpainting · BrushBench | ImageReward x10^2: 38.40 -> 53.86 | Improved |
| replace SFT with RL (GRPO) training | Flux.1-fill · FluxBench | ImageReward x10^2: -34.57 -> 31.41 | Improved |
| replace SFT with RL (GRPO) training | SD1.5-inpainting · BrushBench | ImageReward x10^2: 38.81 -> 53.86 | Improved |
| RL without ground-truth reconstruction constraint | Flux.1-fill · FluxBench | PSNR (mask preservation): 6.993 -> 6.278 | Hurt |
| add boundary reward (BGCS) with fixed weights (ablation) | Flux.1-fill RL · FluxBench | BGCS: -0.0304 -> -0.0183 | Improved |
| add dynamic mask-ratio weighting to the reward (ablation) | Flux.1-fill RL · FluxBench | BGCS: -0.0183 -> -0.0163 | Improved |
| add dynamic mask-ratio weighting to the reward (ablation) | Flux.1-fill RL · FluxBench | HPSv3: 4.193 -> 4.461 | Improved |
| add dynamic weighting and boundary reward (ablation) | Flux.1-fill RL · BrushBench | BGCS: -0.1232 -> -0.1114 | Improved |
| RL with only quality and text rewards, no boundary term (ablation) | Flux.1-fill RL · FluxBench | BGCS: -0.0369 -> -0.0304 | No clear effect |
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.
- PrefPaint — PrefPaint: Aligning Image Inpainting Diffusion Model with Human Preference (NeurIPS 2024) · better on 3 of 3 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
- PrefPaint: Aligning Image Inpainting Diffusion Model with Human Preference (NeurIPS 2024 · similarity 0.92)
- Multimodal Large Language Models Make Text-to-Image Generative Models Align Better (NeurIPS 2024 · similarity 0.87)
- Aesthetic Post-Training Diffusion Models from Generic Preferences with Step-by-step Preference Optimization (CVPR 2025 · similarity 0.87)
- Personalized Image Editing in Text-to-Image Diffusion Models via Collaborative Direct Preference Optimization (NeurIPS 2025 · similarity 0.87)
- Reinforcement Learning Meets Masked Generative Models: Mask-GRPO for Text-to-Image Generation (NeurIPS 2025 · similarity 0.87)
- HD-Painter: High-Resolution and Prompt-Faithful Text-Guided Image Inpainting with Diffusion Models (ICLR 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.
