Research map / Image generation / Text-to-image diffusion
POCA: Pareto-Optimal Curriculum Alignment for Visual Text Generation
Yaohou Fan, Qingzhong Wang, Yongsong Huang, Junyi Liu, Tomo Miyazaki, Shinichiro Omachi
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 |
|---|---|---|---|
| replace weighted-sum rewards with Pareto-guided selection (ablation) | GRPO baseline · AnyText-benchmark English | Sen.ACC: 0.7246 -> 0.7378 | Improved |
| replace weighted-sum rewards with Pareto-guided selection (ablation) | GRPO baseline · AnyText-benchmark Chinese | Sen.ACC: 0.6782 -> 0.6867 | Improved |
| use bi-directional instead of non-dominated-only Pareto sorting (ablation) | Pareto-guided GRPO · AnyText-benchmark English | Sen.ACC: 0.7227 -> 0.7378 | Improved |
| use bi-directional instead of fully-dominated-only sorting (ablation) | Pareto-guided GRPO · AnyText-benchmark English | Sen.ACC: 0.7274 -> 0.7378 | Improved |
| use bi-directional sorting in best@8 (ablation) | Pareto-guided GRPO · AnyText-benchmark English best@8 | Sen.ACC: 0.8418 -> 0.8491 | Improved |
| add adaptive curriculum planning (10k prompts) (ablation) | Pareto-guided GRPO · AnyText-benchmark English | Sen.ACC: 0.7378 -> 0.7651 | Improved |
| increase curriculum data from 1k to 10k prompts (ablation) | POCA · AnyText-benchmark English | Sen.ACC: 0.7374 -> 0.7651 | Improved |
| use 1k-prompt curriculum instead of no curriculum (ablation) | Pareto-guided GRPO · AnyText-benchmark English | Sen.ACC: 0.7378 vs 0.7374 | No clear effect |
| use multi-stage curriculum instead of one stage (ablation) | POCA · AnyText-benchmark English | Sen.ACC: 0.7572 -> 0.7651 | Improved |
| add bi-directional Pareto sorting (ablation) | Glyph-SDXL-v2 · AnyText-benchmark English | Sen.ACC: 0.5950 -> 0.6119 | Improved |
| add curriculum on top of Pareto-guided (ablation) | Glyph-SDXL-v2 · AnyText-benchmark English | Sen.ACC: 0.6119 -> 0.6218 | Improved |
| add POCA to AnyText | AnyText · AnyText-benchmark English | Sen.ACC: 0.7041 -> 0.7651 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- ControlNet — ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback (ECCV 2024) · better on 2 of 2 shared results
- GlyphDraw2 — GlyphDraw2: Automatic Generation of Complex Glyph Posters with Diffusion Models and Large Language Models (AAAI 2025) · better on 1 of 1 shared result
- TextDiffuser — TextDiffuser: Diffusion Models as Text Painters (NeurIPS 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Curriculum Direct Preference Optimization for Diffusion and Consistency Models (CVPR 2025 · similarity 0.87)
- Discriminative Probing and Tuning for Text-to-Image Generation (CVPR 2024 · similarity 0.86)
- Policy Optimized Text-to-Image Pipeline Design (NeurIPS 2025 · similarity 0.86)
- OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation (NeurIPS 2025 · similarity 0.86)
- Learning to Generate via Understanding: Understanding-Driven Intrinsic Rewarding for Unified Multimodal Models (CVPR 2026 · similarity 0.85)
- Seeing is Improving: Visual Feedback for Iterative Text Layout Refinement (CVPR 2026 · similarity 0.85)
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.
