当漂亮不再有用：探究现代文本到图像模型为何无法作为可靠的训练数据生成器
When Pretty Isn't Useful: Investigating Why Modern Text-to-Image Models Fail as Reliable Training Data Generators
对做合成数据训练、数据增强或生成式数据引擎的视觉研究者，这篇论文提示：T2I 模型的视觉保真度与提示词遵循度提升并不等于训练数据可用性提升，需在数据生成流程中显式度量生成分布的多样性与对真实分布的覆盖度，而非只看 FID 或人工观感。
原文依据 classification accuracy on real test data consistently declines with newer T2I models as training data generators
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近期的文本到图像（T2I）扩散模型能够生成视觉上惊艳的图像，并展现出优秀的提示词遵循能力。但它们在作为合成视觉数据生成器时表现如何？在本工作中，我们重新审视了合成数据作为真实训练集可扩展替代品的这一前景，并揭示出一个令人意外的性能退化现象。我们使用 2022 至 2025 年间发布的最优 T2I 模型生成大规模合成数据集，仅在该合成数据上训练标准分类器，并在真实测试数据上进行评估。尽管在视觉保真度和提示词遵循度上可以观察到明显进步，但作为训练数据生成器，越新的 T2I 模型所训练出的分类器在真实测试数据上的分类准确率却持续下降。我们的分析揭示出一个隐藏的趋势：这些模型坍缩到一个狭窄的、以美学为中心的分布，从而削弱了多样性与对真实数据分布的覆盖。总体而言，我们的发现挑战了视觉研究中一个日益流行的假设，即生成真实感的进步意味着数据真实感的进步。因此，我们强调亟需重新思考现代 T2I 模型作为可靠训练数据生成器的能力。
英文原文摘要 1158 字符
Recent text-to-image (T2I) diffusion models produce visually stunning images and demonstrate excellent prompt following. But do they perform well as synthetic vision data generators? In this work, we revisit the promise of synthetic data as a scalable substitute for real training sets and uncover a surprising performance regression. We generate large-scale synthetic datasets using state-of-the-art T2I models released between 2022 and 2025, train standard classifiers solely on this synthetic data, and evaluate them on real test data. Despite observable advances in visual fidelity and prompt adherence, classification accuracy on real test data consistently declines with newer T2I models as training data generators. Our analysis reveals a hidden trend: These models collapse to a narrow, aesthetic-centric distribution that undermines diversity and real data distribution coverage. Overall, our findings challenge a growing assumption in vision research, namely that progress in generative realism implies progress in data realism. We thus highlight an urgent need to rethink the capabilities of modern T2I models as reliable training data generators.