Research map / Image generation / Text-to-image diffusion
Personalized Image Generation via Human-in-the-loop Bayesian Optimization
Rajalaxmi Rajagopalan, Debottam Dutta, Yu-Lin Wei, Romit Roy Choudhury
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 multi-choice (multiwise) preference likelihood instead of pairwise (ablation) | MultiBO · target alignment | CLIP-I: 0.81 -> 0.94 | Improved |
| use multi-choice (multiwise) preference likelihood instead of pairwise (ablation) | MultiBO · target alignment | LPIPS (lower better): 0.68 -> 0.55 | Improved |
| increase choice set size from K=4 to K=6 (ablation) | MultiBO · target alignment | LPIPS (lower better): 0.55 -> 0.54 | No clear effect |
| increase choice set size from K=4 to K=10 (ablation) | MultiBO · target alignment | LPIPS (lower better): 0.55 -> 0.64 | Hurt |
| optimize in attention space instead of 100D latent subspace of x_t (ablation) | MultiBO · target alignment | CLIP-I: 0.8014 -> 0.9364 | Improved |
| optimize in attention space instead of 100D latent subspace of x_t (ablation) | MultiBO · target alignment | LPIPS (lower better): 0.6726 -> 0.5497 | Improved |
| use human scorer instead of CLIP-I2I metric scorer (ablation) | MultiBO · target alignment | CLIP-I: 0.9246 -> 0.9364 | Improved |
| use human scorer instead of LPIPS metric scorer (ablation) | MultiBO · target alignment | LPIPS (lower better): 0.5924 -> 0.5497 | Improved |
| edit middle to later self-attention layers (34-64) rather than layers 1-24 (ablation) | MultiBO · SDXL attention layer choice | qualitative | Improved |
| edit attention only in first 20% of steps rather than whole trajectory (ablation) | MultiBO · timestep range | qualitative | Improved |
| use multi-choice human BO instead of DEMON choose generate mode | DEMON_choose · human evaluation | win rate (%): 15.60 -> 70.82 | Improved |
| use MultiBO with human scorer instead of Aesthetic reward scorer | MultiBO_Aesthetic · human evaluation | win rate (%): 11.26 -> 70.82 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- IterComp — IterComp: Iterative Composition-Aware Feedback Learning from Model Gallery for Text-to-Image Generation (ICLR 2025) · better on 3 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · similarity 0.89)
- Devil is in the Detail: Towards Injecting Fine Details of Image Prompt in Image Generation via Conflict-free Guidance and Stratified Attention (CVPR 2025 · similarity 0.87)
- David and Goliath: Small One-step Model Beats Large Diffusion with Score Post-training (ICML 2025 · similarity 0.87)
- Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation (NeurIPS 2025 · similarity 0.87)
- A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis (AAAI 2024 · similarity 0.87)
- GenIR: Generative Visual Feedback for Mental Image Retrieval (NeurIPS 2025 · 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.
