为4亿张图像标注性别种族，揭示数据偏见如何影响模型
Person-Centric Annotations of LAION-400M: Auditing Bias and Its Transfer to Models
- 用自动管道对LAION-400M图像中人物进行性别种族标注
- 发现男性及非裔、中东裔群体被过度关联负面内容
- 证明数据共现可解释CLIP与Stable Diffusion60%-70%的性别偏见
基于大规模多模态数据集训练的视觉语言模型存在显著人口统计学偏见，但其成因尚不明确。主要障碍在于网页级数据集（如LAION-400M）缺乏人口属性标注。本文通过构建全量人物中心标注，包含超过2.76亿个边界框、感知性别与种族/族裔标签，以及自动生成的描述文本。这些标注基于经验证的自动标注流程，结合目标检测、多模态描述生成和微调分类器实现。利用该资源，我们发现人口不平衡与有害关联，例如男性及被感知为黑人或中东裔个体被不成比例地与犯罪及负面内容关联。此外，线性拟合表明，仅从数据中直接共现即可预测60%-70%的CLIP和Stable Diffusion的性别偏见。本工作建立了数据构成与下游模型偏见之间的首个大规模实证联系。代码已开源：https://github.com/ExplainableML/LAION-400M-Person-Centric-Annotations。
原文摘要 · Abstract (English)
Vision-language models trained on large-scale multimodal datasets show strong demographic biases, but the role of training data in producing these biases remains unclear. A major barrier has been the lack of demographic annotations in web-scale datasets such as LAION-400M. We address this gap by creating person-centric annotations for the full dataset, including over 276 million bounding boxes, perceived gender and race/ethnicity labels, and automatically generated captions. These annotations are produced through validated automatic labeling pipelines combining object detection, multimodal captioning, and finetuned classifiers. Using them, we uncover demographic imbalances and harmful associations, such as the disproportionate linking of men and individuals perceived as Black or Middle Eastern with crime-related and negative content. We also show that a linear fit predicts 60-70% of gender bias in CLIP and Stable Diffusion from direct co-occurrences in the data. Our resources establish the first large-scale empirical link between dataset composition and downstream model bias. Code is available at https://github.com/ExplainableML/LAION-400M-Person-Centric-Annotations.
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。