Research map / Image generation / Text-to-image diffusion
Curriculum Group Policy Optimization: Adaptive Sampling for Unleashing the Potential of Text-to-Image Generation
Baoteng Li, Xianghao Zang, Xinran Wang, Xiangyu Na, Zhixiang He, Hao Sun, Chi Zhang, Zhongjiang He, Tianwei Cao, Kongming Liang, Zhanyu Ma
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 |
|---|---|---|---|
| add variance-based probability sampling (ablation) | Flow-GRPO · GenEval | overall (%): 94.42 -> 95.15 | Improved |
| add exploration balancing (ablation) | CGPO · GenEval | overall (%): 95.15 -> 95.74 | Improved |
| add category calibration (ablation) | CGPO · GenEval | overall (%): 95.74 -> 96.10 | Improved |
| replace Flow-GRPO uniform sampling with CGPO adaptive sampling (efficiency) | Flow-GRPO · GenEval | GPU hours to 0.944: 160 GPU hours, twice the speed | Improved |
| use reward variance instead of advantage magnitude as proxy (ablation) | CGPO · GenEval | overall score: 0.80 -> 0.83 | Improved |
| use reward variance instead of multi-criteria proxy (ablation) | CGPO · GenEval | overall score: 0.82 -> 0.83 | Improved |
| set calibration lambda to 10 instead of 5 (ablation) | Category calibration · GenEval | overall score: 0.95 -> 0.96 | Improved |
| set calibration lambda to 8 instead of 10 (ablation) | Category calibration · GenEval | overall score: 0.96 -> 0.94 | Hurt |
| set calibration lambda to 15 instead of 10 (ablation) | Category calibration · GenEval | overall score: 0.96 -> 0.94 | Hurt |
| replace Flow-GRPO with CGPO | Flow-GRPO · GenEval overall | score: 0.94 -> 0.96 | Improved |
| replace Flow-GRPO with CGPO | Flow-GRPO · GenEval attribute binding | score: 0.82 -> 0.89 | Improved |
| replace Flow-GRPO with CGPO | Flow-GRPO · T2I-CompBench++ Texture | score: 0.7298 -> 0.7521 | Improved |
Closest papers by meaning
- Curriculum Direct Preference Optimization for Diffusion and Consistency Models (CVPR 2025 · similarity 0.89)
- Reinforcement Learning Meets Masked Generative Models: Mask-GRPO for Text-to-Image Generation (NeurIPS 2025 · similarity 0.89)
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 2025 · similarity 0.88)
- Policy Optimized Text-to-Image Pipeline Design (NeurIPS 2025 · similarity 0.87)
- Learning to Sample Effective and Diverse Prompts for Text-to-Image Generation (CVPR 2025 · similarity 0.87)
- Sketch Then Paint: Hierarchical Reinforcement Learning for Diffusion Multi-Modal Large Language Models (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.
