Research map / Image generation / Text-to-image diffusion
Taming Preference Mode Collapse via Directional Decoupling Alignment in Diffusion Reinforcement Learning
Chubin Chen, Sujie Hu, Jiashu Zhu, Meiqi Wu, Jintao Chen, Yanxun Li, Nisha Huang, Chengyu Fang, Jiahong Wu, Xiangxiang Chu, Xiu Li
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 learned b_v correction to reward (ablation) | DanceGRPO · DivGenBench (reward HPS-v2.1) | IDS (lower better): 0.348 -> 0.278 | Improved |
| add learned b_v correction to reward (ablation) | DanceGRPO · DivGenBench (reward HPS-v2.1) | ASC: 0.130 -> 0.205 | Improved |
| add learned b_v correction to reward (ablation) | DanceGRPO · DivGenBench (reward HPS-v2.1) | SDI: 0.488 -> 0.604 | Improved |
| add learned b_v correction to reward (ablation) | DanceGRPO · DivGenBench (reward HPS-v2.1) | PVS: 0.259 -> 0.437 | Improved |
| add learned b_v correction to reward (ablation) | DanceGRPO · HPDv2 (reward HPS-v2.1) | Aesthetic: 6.068 -> 6.353 | Improved |
| add learned b_v correction to reward (ablation) | DanceGRPO · HPDv2 (reward HPS-v2.1) | CLIP: 0.293 -> 0.311 | Improved |
| add learned b_v correction to reward (ablation) | DanceGRPO · HPDv2 (reward HPS-v2.1) | GenEval: 0.522 -> 0.641 | Improved |
| add learned b_v correction to reward (ablation) | DanceGRPO · HPDv2 (reward HPS-v2.1) | HPS-v2.1: 0.361 -> 0.319 | Hurt |
| use learned continuous vector instead of hand-picked discrete words (ablation) | D2-Align · HPDv2 | multiple metrics | Improved |
| use learned b_v instead of uncorrected reward (ablation) | D2-Align · HPDv2 | multiple metrics | Improved |
| train b_v for about 2000 steps or more (ablation) | D2-Align · HPDv2 | performance | Improved |
| set guidance scale to 1.5 (ablation) | D2-Align · HPDv2 | HPS-v2.1 and PickScore | Improved |
Closest papers by meaning
- A Dense Reward View on Aligning Text-to-Image Diffusion with Preference (ICML 2024 · similarity 0.91)
- Confronting Reward Overoptimization for Diffusion Models: A Perspective of Inductive and Primacy Biases (ICML 2024 · similarity 0.90)
- Diffusion Model Alignment Using Direct Preference Optimization (CVPR 2024 · similarity 0.90)
- Calibrated Multi-Preference Optimization for Aligning Diffusion Models (CVPR 2025 · similarity 0.90)
- InPO: Inversion Preference Optimization with Reparametrized DDIM for Efficient Diffusion Model Alignment (CVPR 2025 · similarity 0.90)
- AlignDiff: Aligning Diverse Human Preferences via Behavior-Customisable Diffusion Model (ICLR 2024 · similarity 0.90)
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.
