Research map / Image generation / Text-to-image diffusion
Agentic Retoucher for Text-To-Image Generation
Shaocheng Shen, Jianfeng Liang, Chunlei Cai, Cong Geng, Huiyu Duan, Xiaoyun Zhang, Qiang Hu, Guangtao Zhai
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 lightweight attention module (ablation) | perception agent · GenBlemish-27K saliency | SIM: 0.3731 -> 0.3822 | Improved |
| add lightweight attention module (ablation) | perception agent · GenBlemish-27K saliency | NSS: 1.2153 -> 1.2087 | Hurt |
| add KLD loss (ablation) | perception agent · GenBlemish-27K saliency | SIM: 0.3525 -> 0.3822 | Improved |
| add KLD loss (ablation) | perception agent · GenBlemish-27K saliency | KLD: 1.5008 -> 1.4313 | Improved |
| add GRPO after SFT (progressive alignment) (ablation) | Qwen2.5-VL-7B reasoning agent · GenBlemish-27K reasoning | accuracy (%): 78.34% -> 80.10% | Improved |
| add GRPO after SFT (progressive alignment) (ablation) | GLM-4.1V-9B reasoning agent · GenBlemish-27K reasoning | accuracy (%): 77.13% -> 79.26% | Improved |
| add GRPO after SFT (progressive alignment) (ablation) | Ovis2.5-9B reasoning agent · GenBlemish-27K reasoning | accuracy (%): 78.85% -> 80.62% | Improved |
| add SFT before GRPO (ablation) | Qwen2.5-VL-7B reasoning agent · GenBlemish-27K reasoning | accuracy (%): 58.97% -> 80.10% | Improved |
| apply GRPO alone to the base VLM | Qwen2.5-VL-7B · GenBlemish-27K reasoning | accuracy (%): 57.76% -> 58.97% | Improved |
| add Agentic Retoucher on top of Qwen-Edit | Qwen-Edit · GenBlemish-27K | plausibility: 44.44 -> 47.10 | Improved |
| add Agentic Retoucher on top of SD-inpainting | SD-inpainting · GenBlemish-27K | plausibility: 45.18 -> 46.71 | Improved |
| add Agentic Retoucher on top of Flux-fill | Flux-fill · SynArtifacts-1K | overall: 55.71 -> 57.86 | Improved |
Closest papers by meaning
- JarvisArt: Liberating Human Artistic Creativity via an Intelligent Photo Retouching Agent (NeurIPS 2025 · similarity 0.88)
- Proactive Agents for Multi-Turn Text-to-Image Generation Under Uncertainty (ICML 2025 · similarity 0.87)
- VersaGen: Unleashing Versatile Visual Control for Text-to-Image Synthesis (AAAI 2025 · similarity 0.87)
- Focus-N-Fix: Region-Aware Fine-Tuning for Text-to-Image Generation (CVPR 2025 · similarity 0.87)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · similarity 0.87)
- RetouchIQ: MLLM Agents for Instruction-Based Image Retouching with Generalist Reward (CVPR 2026 · 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.
