small-models-for-glam/index-card-blank-content

该数据集包含裁剪后的单个档案索引卡片，标注为“空白”、“内容”或“分隔符”，用于训练一个轻量级CPU预过滤器，以在卡片目录数字化流程中昂贵的VLM元数据提取之前跳过空白或分隔符卡片。数据集来源于两个集合：波士顿公共图书馆的FRC书架列表卡片和苏格兰国家图书馆的Advocates Library卡片，两个集合的样式不同，因此需要按集合进行评估。数据集通过AI引导和代理验证生成，无需从头开始手动标注，融合了多种弱信号（如NuExtract3 card_type、墨水密度、YOLO卡片检测器框计数等）来生成标签，并通过人工验证黄金测试集以确保泛化能力。数据组成包括训练、验证和测试分割，具体数量在README的表格中列出。数据集适用于图像分类任务，特别用于文档分类和OCR预过滤，但存在一些限制，例如NLS集合在v1版本中仅包含内容卡片，空白卡片尚未包含。

Cropped single archival index cards labelled blank, content, or divider, for training a tiny CPU pre-filter that skips blank/divider cards before expensive VLM metadata extraction in card-catalogue digitisation pipelines. Two collections: Boston Public Library (BPL) FRC shelf-list cards and National Library of Scotland (NLS) Advocates Library cards. Styles differ, so evaluate per collection. The dataset was created through AI-bootstrapped and agent-verified methods, fusing weak signals like NuExtract3 card_type, ink-density with punch-hole removal, and YOLO card-detector box-count, with human verification for disagreements and a gold test split. It includes train, validation, and test splits with specific counts per collection and label. Intended for image classification tasks, particularly document classification and OCR pre-filtering, but with limitations such as NLS contributing only content cards in v1 and blank cards including punch-holes or smudges for model learning.