WildReceipt

WildReceipt是收据的集合。对于每张照片，它包含一个OCR列表-包含边界框，文本和类。
它包含1765张照片，25个类别和50000个文本框。目标是对 “关键信息提取” 进行基准测试-从文档中提取关键信息。有两种不同的模式-文本和视觉特征-这是一个有趣的问题。潜在用途-从文档中提取信息。

WildReceipt is a collection of receipt images. For each image, the dataset provides an OCR list that encompasses bounding boxes, corresponding text segments and category labels. It contains 1765 images, 25 distinct categories and 50,000 text bounding boxes in total. The core objective of this dataset is to benchmark key information extraction (KIE) tasks, which refer to extracting critical information from documents. It incorporates two distinct modalities: textual and visual features, which presents an intriguing research problem. Potential applications include extracting information from documents.

搜集汇总

背景与挑战

背景概述

WildReceipt是一个公开的小型数据集（679.0B），由用户molioosh创建，但未提供具体内容和应用场景描述。