Research map / Image generation / Text-to-image diffusion
Improving Text-to-Image Generation with Intrinsic Self-Confidence Rewards
Seungwook Kim, Minsu Cho
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 |
|---|---|---|---|
| post-train with SOLACE self-confidence reward | SD3.5-M · GenEval | score: 0.65 -> 0.71 | Improved |
| post-train with SOLACE | SD3.5-M · OCR | score: 0.61 -> 0.67 | Improved |
| post-train with SOLACE | SD3.5-M · Human preference (PickScore, HPSv2.1) | score: 22.34 -> 22.41; 0.279 -> 0.278 | No clear effect |
| compute self-confidence without CFG (ablation) | SOLACE · GenEval | score: 0.68 -> 0.71 | Improved |
| compute self-confidence without CFG (ablation) | SOLACE · OCR | score: 0.59 -> 0.67 | Improved |
| compute self-confidence online with the trained model (ablation) | SOLACE · OCR | score: 0.61 -> 0.67 | Improved |
| use K=8 instead of K=4 noise probes (ablation) | SOLACE · OCR | score: 0.66 -> 0.67 | Improved |
| increase probes from K=8 to K=16 (ablation) | SOLACE · Aesthetic | score: 5.39 -> 5.42 | No clear effect |
| use negative advantages (signed GRPO) instead of positive-only (ablation) | SOLACE · OCR | score: 0.62 -> 0.67 | Improved |
| use group size 16 instead of 32 (ablation) | SOLACE · GenEval | score: 0.61 -> 0.71 | Improved |
| use group size 16 instead of 8 (ablation) | SOLACE · GenEval | score: 0.70 -> 0.71 | Improved |
| use aggregated instead of stepwise reward (ablation) | SOLACE · GenEval | score: 0.67 -> 0.71 | Improved |
Closest papers by meaning
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · similarity 0.88)
- Multimodal Large Language Models Make Text-to-Image Generative Models Align Better (NeurIPS 2024 · similarity 0.88)
- Learning to Generate via Understanding: Understanding-Driven Intrinsic Rewarding for Unified Multimodal Models (CVPR 2026 · similarity 0.87)
- Subject-driven Text-to-Image Generation via Preference-based Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Confidence-aware Reward Optimization for Fine-tuning Text-to-Image Models (ICLR 2024 · similarity 0.87)
- Focus-N-Fix: Region-Aware Fine-Tuning for Text-to-Image Generation (CVPR 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.
