Research map / Image generation / Text-to-image diffusion
ImageGem: In-the-wild Generative Image Interaction Dataset for Generative Model Personalization
Yuanhe Guo, Linxi Xie, Zhuoran Chen, Kangrui Yu, Ryan Po, Guandao Yang, Gordon Wetzstein, Hongyi Wen
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 |
|---|---|---|---|
| fine-tune with ImageGem subset instead of Pick-a-Pic subset | SD1.5 DiffusionDPO · DiffusionDPO OOD prompts (Cars) | HPSv2: 0.2690 -> 0.2741 | Improved |
| fine-tune with ImageGem Dogs instead of Pick-a-pic Dogs | SD1.5 DiffusionDPO · DiffusionDPO OOD prompts (Dogs) | HPSv2: 0.2739 -> 0.2789 | Improved |
| fine-tune with ImageGem Scenery Small instead of Pick-a-pic Scenery | SD1.5 DiffusionDPO · DiffusionDPO OOD prompts (Scenery) | Pick Score: 0.1936 -> 0.1949 | Improved |
| scale up scenery subset from small to large (ablation) | SD1.5 DiffusionDPO · DiffusionDPO OOD prompts (Scenery) | HPSv2: 0.2730 -> 0.2747 | Improved |
| scale up cars subset from small to large (ablation) | SD1.5 DiffusionDPO · DiffusionDPO OOD prompts (Cars) | CLIP Score: 0.3745 -> 0.3710 | Hurt |
| DiffusionDPO fine-tuning on scenery vs original SD1.5 | SD1.5 · DiffusionDPO OOD prompts (Scenery) | CLIP Score: 0.3446 -> 0.3427 | Hurt |
| two-tower user embedding instead of ItemKNN | Image retrieval · Images-1M retrieval | Recall@10000: 0.4705 -> 0.5157 | Improved |
| SASRec self-attention over user sequences instead of ItemKNN | Model retrieval · Models-200K ranking | Recall@10: 0.0773 -> 0.1839 | Improved |
| VLM ranking instead of SASRec | Model recommendation · Model recommendation (20 users) | NDCG@5: 0.2795 -> 0.4981 | Improved |
| VLM ranking instead of ItemKNN | Image recommendation · Image recommendation (20 users) | NDCG@5: 0.7065 -> 0.6745 | Hurt |
| SVD top-1 rank reduction of user LoRA (ablation) | SDXL LoRA · SVD preliminary (10178 SDXL LoRAs) | CLIP similarity to base: 0.7563 -> 0.8114 | No clear effect |
| build W2W space from FF layers only (ablation) | W2W editing · ani-real editing | qualitative editing quality | Hurt |
Closest papers by meaning
- Personalized Preference Fine-tuning of Diffusion Models (CVPR 2025 · similarity 0.89)
- Towards Unified Multi-Modal Personalization: Large Vision-Language Models for Generative Recommendation and Beyond (ICLR 2024 · similarity 0.88)
- Multimodal Large Language Models Make Text-to-Image Generative Models Align Better (NeurIPS 2024 · similarity 0.87)
- JeDi: Joint-Image Diffusion Models for Finetuning-Free Personalized Text-to-Image Generation (CVPR 2024 · similarity 0.87)
- Self-Consuming Generative Models with Curated Data Provably Optimize Human Preferences (NeurIPS 2024 · similarity 0.86)
- Inference-Time Personalized Alignment with a Few User Preference Queries (NeurIPS 2025 · similarity 0.86)
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.
