Research map / Image generation / Text-to-image diffusion
The Image as Its Own Reward: Reinforcement Learning with Adversarial Reward for Image Generation
Weijia Mao, Hao Chen, Zhenheng Yang, Mike Zheng Shou
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 adversarial reward model trained against reference images | Flow-GRPO · PickScore reward | PickScore: 22.82 -> 22.78 | No clear effect |
| add adversarial reward model trained against reference images | Flow-GRPO · OCR reward | OCR accuracy: 0.91 -> 0.91 | No clear effect |
| add adversarial reference-based reward | Flow-GRPO · PickScore / OCR rewards | human win rate (quality, aesthetics): 70% quality (PickScore), 85.3% aesthetics (OCR) | Improved |
| use DINO adversarial reward instead of PickScore reward | Flow-GRPO · human evaluation | quality win rate: 93.5% win rate in quality | Improved |
| use DINO adversarial reward | SD3 · OCR prompts | OCR accuracy: 0.59 -> 0.69 | Improved |
| use DINO adversarial reward | SD3 · GenEval prompts | GenEval: 0.61 -> 0.69 | Improved |
| use DINO adversarial reward | SD3 · PickScore prompts | PickScore: 21.70 -> 21.90 | Improved |
| reduce reference images from 1000 to 200 (ablation) | Adv-GRPO (DINO) · DINO similarity | DINO similarity: 0.619 (1000) vs 0.621 (200) | No clear effect |
| add adversarial reward instead of supervised fine-tuning (ablation) | SD3 · PickScore and OCR | PickScore / OCR accuracy: 21.60 -> 22.78 PickScore; 0.68 -> 0.91 OCR | Improved |
| use adversarial reward instead of KL regularization (ablation) | Flow-GRPO · PickScore and OCR | PickScore / OCR accuracy: 21.84 -> 22.78 PickScore; 0.80 -> 0.91 OCR | Improved |
| use adversarial reward instead of PickScore+OCR multi-reward (ablation) | Flow-GRPO · PickScore | PickScore: 21.60 -> 22.78 | Improved |
| use adversarial reward instead of PickScore+OCR multi-reward (ablation) | Flow-GRPO · OCR | OCR accuracy: 0.91 -> 0.91 | No clear effect |
Closest papers by meaning
- Reward-Instruct: A Reward-Centric Approach to Fast Photo-Realistic Image Generation (NeurIPS 2025 · similarity 0.90)
- Reinforcement Learning Meets Masked Generative Models: Mask-GRPO for Text-to-Image Generation (NeurIPS 2025 · similarity 0.87)
- Delving into RL for Image Generation with CoT: A Study on DPO vs. GRPO (NeurIPS 2025 · similarity 0.87)
- Flow-GRPO: Training Flow Matching Models via Online RL (NeurIPS 2025 · similarity 0.87)
- Ctrl-U: Robust Conditional Image Generation via Uncertainty-aware Reward Modeling (ICLR 2025 · similarity 0.87)
- Learning to Sample Effective and Diverse Prompts for Text-to-Image Generation (CVPR 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.
