Research map / Image generation / Text-to-image diffusion
From Broad Exploration to Stable Synthesis: Entropy-Guided Optimization for Autoregressive Image Generation
Han Song, Yucheng Zhou, Jianbing Shen, Yu Cheng
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 entropy guidance to both CoT and image tokens (EG-GRPO vs T2I-R1) | T2I-R1 (GRPO) · T2I-CompBench Shape | score: 58.67 -> 60.88 | Improved |
| add entropy guidance | T2I-R1 (GRPO) · T2I-CompBench Color | score: 82.58 -> 84.11 | Improved |
| add entropy guidance | T2I-R1 (GRPO) · T2I-CompBench Texture | score: 76.94 -> 77.38 | Improved |
| add entropy guidance | T2I-R1 (GRPO) · WISE Culture | score: 48.00 -> 49.00 | Improved |
| add entropy guidance | T2I-R1 (GRPO) · WISE Spatio-temporal | score: 55.50 -> 56.00 | No clear effect |
| add entropy guidance | T2I-R1 (GRPO) · WISE Science | score: 45.00 -> 46.33 | Improved |
| apply entropy guidance to both CoT and image tokens instead of only CoT tokens (ablation) | EG-GRPO · T2I-CompBench Shape | score: 55.68 -> 60.88 | Improved |
| apply entropy guidance to both CoT and image tokens instead of only image tokens (ablation) | EG-GRPO · T2I-CompBench Color | score: 79.25 -> 84.11 | Improved |
| apply entropy guidance to only one modality instead of none (ablation) | T2I-R1 (GRPO) · T2I-CompBench Color | score: 82.58 -> 79.25 | Hurt |
| apply entropy guidance to only CoT tokens instead of none (ablation) | T2I-R1 (GRPO) · WISE Science | score: 45.00 -> 43.67 | Hurt |
| add EG-GRPO entropy mechanism | T2I-R1 (GRPO) · training step time | seconds: 50.20 -> 50.88 | Hurt |
| add EG-GRPO entropy mechanism | T2I-R1 (GRPO) · training reward at step 100 | reward: 2.1105 -> 2.1411 | Improved |
Closest papers by meaning
- Reinforcement Learning Meets Masked Generative Models: Mask-GRPO for Text-to-Image Generation (NeurIPS 2025 · similarity 0.89)
- MaskSense: Confronting the Visual Exploration Trap in Masked Image Generation (NeurIPS 2026 · similarity 0.89)
- Delving into RL for Image Generation with CoT: A Study on DPO vs. GRPO (NeurIPS 2025 · similarity 0.89)
- T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT (NeurIPS 2025 · similarity 0.88)
- Beyond the Dirac Delta: Mitigating Diversity Collapse in Reinforcement Fine-Tuning for Image Generation (NeurIPS 2026 · similarity 0.88)
- Let's Verify and Reinforce Image Generation Step by Step (CVPR 2025 · similarity 0.88)
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.
