新基准测试揭示文生图模型在复杂指令下的忠实度缺陷，提出依赖感知奖励提升精准生成。
Arena-T2I Hard: Benchmarking and Improving Faithfulness with Dependency-Aware Checklist
- 构建310个真实场景复杂指令的评测集，每条含约30个可拆解的约束条件。
- 顶尖闭源模型在该评测中仅达0.855得分，11系统间差距达33个百分点。
- 设计依赖感知检查表奖励，使模型生成更符合具体约束，兼顾美感与准确性。
忠实度——生成图像与提示词的匹配精度——日益成为文本到图像（T2I）模型实际应用的核心。现有基准多基于简单原子指令，顶级系统已接近完美表现。当T2I模型进入创作流程，用户常提出包含复杂空间关系、风格约束和文本渲染的多维度请求，单一二元视觉语言模型（VLM）评分难以定位模型失败的具体环节。本文提出Arena-T2I Hard，一个由真实竞技场日志提取的310条提示组成的压力测试集，每条提示分解为约30个跨六类（含文本渲染）的真假判断约束。评估中最强闭源系统得分0.855，11系统间性能差距达33个百分点，显示强大区分能力。此外，公开竞技场排名无法预测忠实度，说明整体布拉德利-特里（BT）偏好分更关注美学而非细粒度提示遵循。为此，我们提出依赖感知检查表奖励，将提示分解为有向无环图（DAG）形式的真假问题，并在父节点失败时归零其所有子节点，将忠实度转化为逐约束训练信号。结合通过组解耦归一化（GDPO）实现的BT美学奖励，该方案在SD3.5-Medium与FLUX.1-dev上，于MMRB2成对比较中，优于所有单奖励、朴素加权和或4奖励BT集成基线，实现更优的忠实度-美学平衡。
原文摘要 · Abstract (English)
Faithfulness -- how precisely a generated image aligns with its prompt -- is increasingly central to the real-world utility of text-to-image (T2I) models. Existing faithfulness benchmarks, however, rely on simple atomic instructions, on which top-tier systems already achieve near-perfect scores. As T2I models enter creative workflows, users issue multi-faceted requests combining intricate spatial relationships, stylistic constraints, and complex text rendering. In this setting, a single binary VLM-judge score no longer captures which specific constraints the model fails to satisfy. We introduce Arena-T2I Hard, a 310-prompt stress benchmark drawn from real arena T2I logs, with approximately 30 decomposed yes/no constraints per prompt spanning six categories, including text rendering. The strongest closed-source system we evaluate reaches 0.855 with a 33~pp performance gap across 11 systems, demonstrating substantial discriminative power. Moreover, high public-arena rankings fail to predict faithfulness, confirming that holistic Bradley-Terry (BT) preference scores prioritize aesthetics over fine-grained prompt adherence. We propose a dependency-aware checklist reward that decomposes each prompt into a DAG of yes/no questions and zeroes descendants of failed parents, turning faithfulness into a per-constraint training signal. Combined with a BT aesthetic reward via group-decoupled normalization (GDPO), which standardizes each reward within its rollout group so neither collapses, the recipe attains a strictly better faithfulness-aesthetics trade-off on SD3.5-Medium and FLUX.1-dev under MMRB2 pairwise comparisons than every single-reward, naive weighted-sum, or 4-reward BT-ensemble baseline.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。