Research map / Image generation / Text-to-image diffusion
Towards Fine-Grained Attribution: Instance-Aware Preference Optimization for Aligning Diffusion Models
Jiayang Sun, Pin Wang, Hongbo Wang, Xinyue Liu, Huaibo Huang, 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 |
|---|---|---|---|
| replace image-level DPO with instance-level preference optimization (IAPO) vs InPO | SD1.5 · HPD v2 | Aesthetic (mean): 5.8064 -> 5.9261 | Improved |
| IAPO vs InPO on CLIP score | SD1.5 · HPD v2 | CLIP (mean): 36.4773 -> 36.4029 | Hurt |
| IAPO vs InPO on CLIP score | SDXL · Parti-Prompts | CLIP (mean): 35.6453 -> 35.3439 | Hurt |
| IAPO vs InPO on Pick-a-Pic v2 test | SD1.5 · Pick-a-Pic v2 test | Aesthetic (mean): 5.6566 -> 5.7842 | Improved |
| IAPO vs InPO on SDXL | SDXL · Pick-a-Pic v2 test | Aesthetic (mean): 6.0416 -> 6.1226 | Improved |
| instance-level reweighting makes training faster than InPO | SD1.5 · Pick-a-Pic v2 | training GPU hours (H800): ~57.6 -> ~17.6 | Improved |
| lower w_neg from 1 to 0 (stronger instance-level suppression) (ablation) | IAPO SD1.5 · Pick-a-Pic v2 test | Aesthetic (mean): 5.73 -> 5.78 | Improved |
| lower w_neg from 1 to 0 (ablation) | IAPO SD1.5 · Pick-a-Pic v2 test | PickScore (mean): 21.52 -> 21.58 | Improved |
| lower w_neg from 1 to 0 (ablation) | IAPO SD1.5 · Pick-a-Pic v2 test | HPS (mean): 27.76 -> 27.85 | Improved |
| lower w_neg from 1 to 0 (ablation) | IAPO SD1.5 · Pick-a-Pic v2 test | CLIP (mean): 34.66 -> 34.70 | No clear effect |
Closest papers by meaning
- Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences (ICML 2025 · similarity 0.90)
- Diffusion Model Alignment Using Direct Preference Optimization (CVPR 2024 · similarity 0.90)
- InPO: Inversion Preference Optimization with Reparametrized DDIM for Efficient Diffusion Model Alignment (CVPR 2025 · similarity 0.90)
- Calibrated Multi-Preference Optimization for Aligning Diffusion Models (CVPR 2025 · similarity 0.90)
- Aligning Diffusion Models by Optimizing Human Utility (NeurIPS 2024 · similarity 0.90)
- MIA-DPO: Multi-Image Augmented Direct Preference Optimization For Large Vision-Language Models (ICLR 2025 · 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.
