Research map / Image generation / Text-to-image diffusion
CREward: A Type-Specific Creativity Reward Model
Jiyeon Han, Ali Mahdavi-Amiri, Hao Zhang, Haedong Jeong
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 SigLIP vision backbone instead of VGG16 for reward model (ablation) | CREward · LVLM-labelled test pairs, overall | test preference accuracy: 0.74 -> 0.82 | Improved |
| use SigLIP instead of DINOv3 backbone (ablation) | CREward · LVLM-labelled test pairs, material | test preference accuracy: 0.70 -> 0.78 | Improved |
| use larger backbone (DINOv3, 824M) instead of CLIP (304M) (ablation) | CREward · LVLM-labelled test pairs, overall | test preference accuracy: 0.78 -> 0.78 | No clear effect |
| use Gemini-2.5 Flash as annotator instead of Gemma-3 | LVLM creativity labelling · CREBench-Human geometry | rank correlation: 0.56 -> 0.80 | Improved |
| distill Gemma-3 labels into CREward versus using Gemma-3 itself | creativity scoring · CREBench-Human texture | rank correlation: 0.60 -> 0.76 | Improved |
| distill Gemma-3 labels into CREward versus using Gemma-3 itself | creativity scoring · CREBench-Human geometry | rank correlation: 0.56 -> 0.59 | Mixed |
| use type-specific CREward instead of type-agnostic Surprise score | creativity scoring · CREBench-Human overall | rank correlation: 0.56 -> 0.61 | Improved |
| apply C3 to SDXL | SDXL · CREward reward scores | creativity score | Improved |
| train LoRA sliders with CREward reward | T2I diffusion model · qualitative generation | type-specific creativity | Improved |
Closest papers by meaning
- CreBench: Human-Aligned Creativity Evaluation from Idea to Process to Product (AAAI 2026 · similarity 0.86)
- Redefining <Creative> in Dictionary: Towards an Enhanced Semantic Understanding of Creative Generation (CVPR 2025 · similarity 0.84)
- Social Reward: Evaluating and Enhancing Generative AI through Million-User Feedback from an Online Creative Community (ICLR 2024 · similarity 0.84)
- IDEA-Bench: How Far are Generative Models from Professional Designing? (CVPR 2025 · similarity 0.84)
- An analytic theory of creativity in convolutional diffusion models (ICML 2025 · similarity 0.84)
- Enhancing Creative Generation on Stable Diffusion-based Models (CVPR 2025 · similarity 0.83)
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.
