Research map / Image generation / Text-to-image diffusion
Diffusion-SDPO: Safeguarded Direct Preference Optimization for Diffusion Models
Minghao Fu, Guo-Hua Wang, Tianyu Cui, Qing-Guo Chen, Zhao Xu, Weihua Luo, Kaifu Zhang
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 SDPO safeguard | Diff.-DPO · HPS V2 prompts (SD 1.5) | PickScore: 0.2131 -> 0.2174 | Improved |
| add SDPO safeguard | Diff.-DPO · HPS V2 prompts (SD 1.5) | ImageReward: 0.1705 -> 0.6211 | Improved |
| add SDPO safeguard | DMPO · HPS V2 prompts (SD 1.5) | HPS V2: 0.2766 -> 0.2848 | Improved |
| add SDPO safeguard | DSPO · HPS V2 prompts (SD 1.5) | HPS V2: 0.2837 -> 0.2847 | Improved |
| add SDPO safeguard | DSPO · HPS V2 prompts (SD 1.5) | CLIP: 0.3598 -> 0.3586 | Hurt |
| add SDPO safeguard | Diff.-DPO · HPS V2 prompts (SDXL) | Aesthetics: 6.1380 -> 6.1284 | Hurt |
| add SDPO safeguard | DSPO · HPS V2 prompts (SDXL) | ImageReward: 1.0054 -> 1.0745 | Improved |
| add SDPO safeguard | Diff.-DPO · SD 1.5 win rate vs base (mean) | win rate (%): 67.90 | Improved |
| add SDPO safeguard | DSPO · SD 1.5 win rate vs base (mean) | win rate (%): 51.95 | No clear effect |
| replace DPO with SDPO | Ovis-U1 · DPG-Bench | score: 83.78 -> 84.84 | Improved |
| replace DPO with SDPO | Ovis-U1 · ImgEdit | score: 4.01 -> 4.11 | Improved |
| add SDPO safeguard | Diff.-DPO · FLUX.1-dev GenEval | GenEval: 0.23 -> 0.71 | Improved |
Closest papers by meaning
- Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences (ICML 2025 · similarity 0.94)
- InPO: Inversion Preference Optimization with Reparametrized DDIM for Efficient Diffusion Model Alignment (CVPR 2025 · similarity 0.92)
- Diffusion Model Alignment Using Direct Preference Optimization (CVPR 2024 · similarity 0.92)
- Aesthetic Post-Training Diffusion Models from Generic Preferences with Step-by-step Preference Optimization (CVPR 2025 · similarity 0.92)
- Ranking-based Preference Optimization for Diffusion Models from Implicit User Feedback (NeurIPS 2025 · similarity 0.92)
- Calibrated Multi-Preference Optimization for Aligning Diffusion Models (CVPR 2025 · similarity 0.92)
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.
