合成发票矫正数据集

本数据集由SyntheticInvoice_train10k,SyntheticInvoice_test1k,SyntheticInvoice_val1k三部分构成，将已有的电子发票旋转任意角度后随机放置在任意背景上，背景数据来自于(DTD)。训练数据1万张，测试和验证数据各1000张。

This dataset comprises three subsets: SyntheticInvoice_train10k, SyntheticInvoice_test1k, and SyntheticInvoice_val1k. All samples are generated by randomly rotating pre-existing electronic invoices at arbitrary angles and randomly placing them onto arbitrary backgrounds, where the background data is sourced from the Describable Textures Dataset (DTD). Specifically, there are 10,000 training samples, and 1,000 samples each for the test and validation subsets.

搜集汇总

背景与挑战

背景概述

该数据集通过合成开源发票和纹理图像生成通用发票数据，用于训练发票检测、定位和识别模型，避免敏感信息泄露。它包括10,000张训练图像和1,000张测试及验证图像，支持发票检测、矫正和识别任务，并需遵守相关开源许可协议。