# Decompose Radicals, Then Reward: Fine-Grained Inspection for Accurate Chinese Text Rendering

arXiv: https://arxiv.org/abs/2609.37569 · HTML: https://mlsanity.com/paper/2609.37569/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 27 of its 32 references that are in the map.
- First posted on arXiv 29 Sep 2026.

## Builds on (most cited first, 27 of 27)

- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen-Image Technical Report](https://mlsanity.com/paper/2508.02324/) (2025)
- [Flow-GRPO: Training Flow Matching Models via Online RL](https://mlsanity.com/paper/2505.05470/) (2025)
- [TextDiffuser: Diffusion Models as Text Painters](https://mlsanity.com/paper/2305.10855/) (2023)
- [AnyText: Multilingual Visual Text Generation And Editing](https://mlsanity.com/paper/2311.03054/) (2023)
- [PaddleOCR 3.0 Technical Report](https://mlsanity.com/paper/2507.05595/) (2025)
- [TextDiffuser-2: Unleashing the Power of Language Models for Text Rendering](https://mlsanity.com/paper/2311.16465/) (2023)
- [X-Omni: Reinforcement Learning Makes Discrete Autoregressive Image Generative Models Great Again](https://mlsanity.com/paper/2507.22058/) (2025)
- [NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognition](https://mlsanity.com/paper/1806.00926/) (2018)
- [AnyText2: Visual Text Generation and Editing With Customizable Attributes](https://mlsanity.com/paper/2411.15245/) (2024)
- [GlyphDraw2: Automatic Generation of Complex Glyph Posters with Diffusion Models and Large Language Models](https://mlsanity.com/paper/2407.02252/) (2024)
- [Hi3D: Pursuing High-Resolution Image-to-3D Generation with Video Diffusion Models](https://mlsanity.com/paper/2409.07452/) (2024)
- [Zero-Shot Chinese Character Recognition with Stroke-Level Decomposition](https://mlsanity.com/paper/2106.11613/) (2021)
- [Chinese Text Recognition with A Pre-Trained CLIP-Like Model Through Image-IDS Aligning](https://mlsanity.com/paper/2309.01083/) (2023)
- [SVTRv2: CTC Beats Encoder-Decoder Models in Scene Text Recognition](https://mlsanity.com/paper/2411.15858/) (2024)
- [3DStyle-Diffusion: Pursuing Fine-grained Text-driven 3D Stylization with 2D Diffusion Models](https://mlsanity.com/paper/2311.05464/) (2023)
- [TextPecker: Rewarding Structural Anomaly Quantification for Enhancing Visual Text Rendering](https://mlsanity.com/paper/2602.20903/) (2026)
- [HunyuanOCR-1.5: Making Lightweight OCR VLMs Faster and Better](https://mlsanity.com/paper/2607.04884/) (2026)
- [Flow-Factory: A Unified Framework for Reinforcement Learning in Flow-Matching Models](https://mlsanity.com/paper/2602.12529/) (2026)
- [OCRGenBench: A Comprehensive Benchmark for Evaluating OCR Generative Capabilities](https://mlsanity.com/paper/2507.15085/) (2025)
- [TextSSR: Diffusion-based Data Synthesis for Scene Text Recognition](https://mlsanity.com/paper/2412.01137/) (2024)
- [GlyphPrinter: Region-Grouped Direct Preference Optimization for Glyph-Accurate Visual Text Rendering](https://mlsanity.com/paper/2603.15616/) (2026)
- [TextAlign: Preference Alignment for Text Rendering with Hierarchical Rewards](https://mlsanity.com/paper/2605.19320/) (2026)
- [Advancing WordArt-Oriented Scene Text Recognition: Datasets and Methods](https://mlsanity.com/paper/2606.24484/) (2026)
- [Zero-Shot Chinese Character Recognition with Hierarchical Multi-Granularity Image-Text Aligning](https://mlsanity.com/paper/2505.24837/) (2025)
- [All-in-One Multilingual Scene Text Recognition with Script-aware Mixture-of-Experts](https://mlsanity.com/paper/2609.24058/) (2026)

## Abstract (arXiv)

Rendering accurate Chinese text remains challenging for text-to-image models. Existing OCR-based reinforcement-learning rewards compare decoded transcripts with target strings. Such rewards overlook the compositional nature of Chinese writing: an ideograph consists of reusable components arranged through explicit spatial relations, yet OCR evaluates it as an atomic character. Consequently, visually different radical-level errors may receive equally coarse feedback, encouraging glyphs that merely resemble the target instead of faithfully reproducing its internal structure. We employ Ideographic Description Sequences (IDS), which comprise spatial operators and character components, and train an expert IDS recognizer to transcribe rendered Chinese text into this representation. Building on this recognizer, we introduce IDSpect, which deterministically decomposes the target text into IDS tokens and aligns crop-level visual IDS predictions with the target sequence. Globally unique token credit makes this comparison robust to the order of detected text regions. Combined with a whole-character semantic reward, IDSpect supplies fine-grained credit with component and spatial-relation without changing the image generator or adding inference-time cost. Experiments with GRPO post-training of Qwen-Image demonstrate that IDSpect achieves leading structural quality and semantic alignment on LongText and GenTextEval.