Blame Errant Words, Not the Canvas: Defect-Aware Spatial Credit Assignment for Visual Text Generation

查看原文 / View Original 编辑 / Edit

Computer Science > Computer Vision and Pattern Recognition arXiv:2610.20794 (cs) [Submitted on 14 Oct 2026] Title: Blame Errant Words, Not the Canvas: Defect-Aware Spatial Credit Assignment for Visual Text Generation Authors: Sujie Hu, Huaisong Zhang, Pujing Yang, Qian Wang, Xiaomin Li, Xinhang Leng, Chen Li, Jing Lyu, Xiu Li, Ling Pan.

计算机科学 > 计算机视觉与模式识别 arXiv:2610.20794 (cs) [提交于 2026 年 10 月 14 日] 标题：责怪错误的文字，而非画布：视觉文本生成的缺陷感知空间信用分配作者：Sujie Hu, Huaisong Zhang, Pujing Yang, Qian Wang, Xiaomin Li, Xinhang Leng, Chen Li, Jing Lyu, Xiu Li, Ling Pan。

Abstract: Recent reinforcement learning approaches have improved visual text generation, yet generated images still suffer from incorrect wording and malformed glyphs. We identify two coupled limitations: recognition-based rewards often overlook visible glyph defects, while whole-image scalar feedback obscures the uneven quality of individual text regions.

摘要：近期的强化学习方法改善了视觉文本生成，但生成的图像仍然存在文字错误和字形畸变的问题。我们确定了两个相互关联的局限性：基于识别的奖励往往忽略了可见的字形缺陷，而全图标量反馈则掩盖了各个文本区域质量的不均衡。

These limitations motivate our central principle: $\textit{Blame errant words, not the canvas}$, to effectively preserve local quality distinctions. First, we introduce $\mathbf{TextDART}$, a defect-aware region-targeted reward model that localizes individual text regions and independently evaluates their semantic correctness and glyph quality, providing fine-grained, region-precise feedback.

这些局限性促成了我们的核心原则：“责怪错误的文字，而非画布”，以有效保留局部质量差异。首先，我们引入了 $\mathbf{TextDART}$，这是一个缺陷感知的区域目标奖励模型，它能够定位单个文本区域并独立评估其语义正确性和字形质量，从而提供细粒度、区域精确的反馈。

Building on TextDART, we propose $\mathbf{SCANNER}$, which turns these regional judgments into spatially differentiated policy feedback. It normalizes rewards for each text target across sampled images and casts the resulting regional advantages onto the corresponding latent tokens.

在 TextDART 的基础上，我们提出了 $\mathbf{SCANNER}$，它将这些区域判断转化为空间差异化的策略反馈。它对采样图像中每个文本目标的奖励进行归一化，并将由此产生的区域优势映射到相应的潜在标记上。

Extensive experiments on existing benchmarks and our newly established $\textbf{PolyText-Bench}$ demonstrate that our approach achieves state-of-the-art performance in text accuracy and glyph fidelity.

在现有基准测试和我们新建立的 $\textbf{PolyText-Bench}$ 上进行的广泛实验表明，我们的方法在文本准确性和字形保真度方面达到了最先进的性能。