Research map / Image generation / Text-to-image diffusion
Beyond Linear Shortcuts: Rectifying Diffusion Preference Optimization with Intrinsic Generative Geometry
Pin Wang, Huaibo Huang, Jiayang Sun, Hongbo Wang, Wentao Jiang, Tiezheng Ge, Jie Cao, Ran He
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 |
|---|---|---|---|
| use TRF fusion weight 0.2 instead of 0.0 (geometric anchor only nullified) (ablation) | TCPO-SDXL · Pick-a-Pic v2 test | HPS (mean): 28.6238 -> 28.8740 | Improved |
| use TRF fusion weight 0.2 instead of 0.4 (ablation) | TCPO-SDXL · Pick-a-Pic v2 test | HPS (mean): 28.5480 -> 28.8740 | Improved |
| use TRF fusion weight 0.2 instead of 0.0 (ablation) | TCPO-SDXL · Pick-a-Pic v2 test | MPS (mean): 10.2044 -> 12.2160 | Improved |
| use SLERP instead of linear interpolation in TRF (ablation) | TCPO-SDXL · Pick-a-Pic v2 test | HPS (mean): 28.8214 -> 28.8740 | Improved |
| use SLERP instead of linear interpolation in TRF (ablation) | TCPO-SDXL · Pick-a-Pic v2 test | Aesthetic (mean): 6.0942 -> 6.0744 | Hurt |
| vary anchor granularity alpha 1.43 / 2.0 / 3.33 (ablation) | TCPO-SDXL · Pick-a-Pic v2 test | HPS (mean): 28.8297 / 28.8740 / 28.7977 | No clear effect |
| add curvature-adaptive intermediate anchor (alpha>1) over endpoint DPO | TCPO-SDXL · Pick-a-Pic v2 test | PickScore (mean): 22.6397 -> 22.8045 | Improved |
| use TCPO instead of InPO | SDXL · PartiPrompts | HPS (mean): 28.8443 -> 29.1819 | Improved |
| use TCPO instead of DPO | SDXL · Pick-a-Pic v2 test | HPS (mean): 28.5933 -> 28.8740 | Improved |
| use TCPO instead of InPO | SDXL · Pick-a-Pic v2 test | PickScore (mean): 22.5820 -> 22.8045 | Improved |
| use TCPO instead of MaPO | SDXL · Pick-a-Pic v2 test | Aesthetic (mean): 6.2037 -> 6.0744 | Hurt |
| fine-tune SDXL with TCPO | SDXL · HPDv2 | PickScore: 22.7835 -> 23.3950 | Improved |
Closest papers by meaning
- Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences (ICML 2025 · similarity 0.91)
- InPO: Inversion Preference Optimization with Reparametrized DDIM for Efficient Diffusion Model Alignment (CVPR 2025 · similarity 0.89)
- Diffusion-NPO: Negative Preference Optimization for Better Preference Aligned Generation of Diffusion Models (ICLR 2025 · similarity 0.88)
- DrPO: Drifting Preference Optimization for One-Step Generative Models (NeurIPS 2026 · similarity 0.88)
- D-Fusion: Direct Preference Optimization for Aligning Diffusion Models with Visually Consistent Samples (ICML 2025 · similarity 0.88)
- TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment (NeurIPS 2026 · 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.
