Research map / Image generation / Text-to-image diffusion
TextGround4M: A Prompt-Aligned Dataset for Layout-Aware Text Rendering
Dongxing Mao, Yilin Wang, Linjie Li, Zhengyuan Yang, Alex Jinpeng Wang
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 text-token supervision after image tokens (ablation) | Janus Pro 1B fine-tuned · TextGround-Bench (Easy) | Prompt Coverage: 15.02 -> 16.35 | Improved |
| add bbox-token supervision only (ablation) | Janus Pro 1B fine-tuned · TextGround-Bench (Easy) | Prompt Coverage: 15.02 -> 30.61 | Improved |
| add bbox-token supervision only (ablation) | Janus Pro 1B fine-tuned · TextGround-Bench (Easy) | Layout IoU: 3.80 -> 7.95 | Improved |
| add text tokens to bbox-only supervision (ablation) | Janus Pro 1B fine-tuned · TextGround-Bench (Mid) | Prompt Coverage: 16.97 -> 19.09 | Improved |
| add text tokens to bbox-only supervision (ablation) | Janus Pro 1B fine-tuned · TextGround-Bench (Easy) | Prompt Coverage: 30.61 -> 30.80 | No clear effect |
| add text tokens to bbox-only supervision (ablation) | Janus Pro 1B fine-tuned · TextGround-Bench (Mid) | Layout IoU: 3.67 -> 4.32 | Improved |
| add text tokens to bbox-only supervision (ablation) | Janus Pro 1B fine-tuned · TextGround-Bench (Hard) | Layout IoU: 0.55 -> 0.80 | Improved |
| place supervision tokens before image instead of after (ablation) | Ours (Post-Image) · TextGround-Bench (Easy) | Prompt Coverage: 30.80 -> 34.41 | Improved |
| place supervision tokens before image instead of after (ablation) | Ours (Post-Image) · TextGround-Bench (Easy) | Layout IoU: 7.93 -> 10.43 | Improved |
| place supervision tokens before image instead of after (ablation) | Ours (Post-Image) · TextGround-Bench (Hard) | Prompt Coverage: 6.09 -> 6.30 | No clear effect |
| place supervision tokens before image instead of after (ablation) | Ours (Post-Image) · TextGround-Bench (Hard) | Layout IoU: 0.80 -> 0.43 | Hurt |
| fine-tune Janus Pro 1B on TextGround4M with layout-aware tokens | Janus Pro 1B · TextGround-Bench (Mid) | Prompt Coverage: 3.33 -> 19.09 | Improved |
Closest papers by meaning
- OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation (NeurIPS 2025 · similarity 0.90)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.89)
- Arena-T2I Hard: Benchmarking and Improving Faithfulness with Dependency-Aware Checklist Rewards (NeurIPS 2026 · similarity 0.89)
- Seeing is Improving: Visual Feedback for Iterative Text Layout Refinement (CVPR 2026 · similarity 0.89)
- TextCenGen: Attention-Guided Text-Centric Background Adaptation for Text-to-Image Generation (ICML 2025 · similarity 0.89)
- Rethinking The Training And Evaluation of Rich-Context Layout-to-Image Generation (NeurIPS 2024 · similarity 0.89)
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.
