small-models-for-glam/index-card-detection-v5

该数据集是Archival Index Card Detection — mixed collections (5版本，集成重新标记海军数据)，是一个用于档案索引卡片检测的混合集合数据集。它是v3版本的改进版本，包含1,425行数据，来自多个来源集合（包括NLS Advocates、Navy Nurse Corps、BPL Catalog和Rubenstein Manuscript）。主要改进在于对25行海军多卡片扫描的边界框进行了重新标记：通过运行v3和v4两个模型，在低置信度阈值下生成检测结果，使用IoU-NMS合并检测，并进行人工审查以修正分歧。最终边界框质量更高，标签来源包括v3+v4一致、v3仅、v4仅或human调整。数据集结构包括图像、来源集合、边界框（xywh格式）、类别（固定为card）和边界框来源等字段。用于训练对象检测器以在全页、多卡片图像或预裁剪单卡片图像中定位档案索引卡片，但仅限于检测任务（不包含OCR）、单类别、英文和英美档案惯例。数据集由Hugging Face的机器学习图书馆员Daniel van Strien策划，许可证混合（依据来源集合不同）。

Refined version of [`small-models-for-glam/index-card-detection-v3`](https://huggingface.co/datasets/small-models-for-glam/index-card-detection-v3). All NLS / BPL / Rubenstein rows are passed through unchanged. The 25 navy-nurse-corps rows have their bounding boxes **re-labelled via a v3+v4 model ensemble plus human review**, replacing the SAM3-only bootstrap from v3. Same 1,425-row mixed-collection composition as v3. The only difference: the 25 navy multi-card scans now use higher-quality bounding boxes produced by: 1. Running both `small-models-for-glam/index-card-detector-v3` and `small-models-for-glam/index-card-detector-v4` on each navy image (per-model conf threshold 0.10). 2. Greedy IoU-NMS at 0.5 to merge near-duplicate detections from the two models, recording which models contributed to each surviving box (`box_source = "v3+v4"`, `"v3"`, or `"v4"`). 3. Human review in a custom HTML bbox editor, prioritising rows where the two models disagreed (rows with `v3`-only or `v4`-only boxes appeared first in the review queue). Outcome: 86 raw ensemble boxes (83 unanimous + 3 disagreement), reduced to 84 after one envelope-style v4 artefact was removed and 2 boxes were structurally adjusted.