Research map / Large language models / Alignment and preferences
Follow-Your-Preference: Towards Preference-Aligned Image Inpainting
Yutao Shen, Junkun Yuan, Toru Aonishi, Hideki Nakayama, Jack 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 |
|---|---|---|---|
| add Ensemble-reward DPO preference alignment (ablation) | BrushNet · BrushBench | GPT-4 score: 79.391 -> 83.054 | Improved |
| add Ensemble-reward DPO preference alignment (ablation) | FLUX.1 Fill · EditBench | GPT-4 score: 66.979 -> 72.307 | Improved |
| add Ensemble-reward alignment on new I Dream My Painting data (ablation) | FLUX.1 Fill · I Dream My Painting | GPT-4 score: 76.391 -> 79.255 | Improved |
| build DPO pairs with PickScore instead of HPSv2 (ablation) | BrushNet · BrushBench | GPT-4 score: 79.914 -> 82.726 | Improved |
| build DPO pairs with HPSv2 instead of PickScore (ablation) | FLUX.1 Fill · BrushBench | GPT-4 score: 84.188 -> 84.699 | Improved |
| build DPO pairs with VQAScore reward (ablation) | BrushNet · BrushBench | GPT-4 score: 79.391 -> 78.877 | Hurt |
| use reward ensemble (average ranking) instead of single reward model (ablation) | preference data construction · BrushBench and EditBench, scaling | multiple metrics | Improved |
| scale candidates and samples with HPSv2 rewards (ablation) | preference data · BrushNet and FLUX.1 Fill | GPT-4 score | Hurt |
| build DPO pairs with random pairs instead of baseline (ablation) | BrushNet · BrushBench | GPT-4 score: 79.391 -> 79.177 | No clear effect |
| scale preference data (candidates and samples) (ablation) | DPO with Ensemble/PickScore · BrushBench and EditBench | reward-model metrics | Improved |
Closest papers by meaning
- PrefPaint: Aligning Image Inpainting Diffusion Model with Human Preference (NeurIPS 2024 · similarity 0.93)
- Diffusion Model as a Noise-Aware Latent Reward Model for Step-Level Preference Optimization (NeurIPS 2025 · similarity 0.89)
- Multimodal Large Language Models Make Text-to-Image Generative Models Align Better (NeurIPS 2024 · similarity 0.88)
- Aligning Visual Contrastive learning models via Preference Optimization (ICLR 2025 · similarity 0.88)
- A Dense Reward View on Aligning Text-to-Image Diffusion with Preference (ICML 2024 · similarity 0.88)
- DreamAlign: Dynamic Text-to-3D Optimization with Human Preference Alignment (AAAI 2025 · similarity 0.88)
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.
