bytedance-research/KIE-HVQA

KIE-HVQA是一个专门设计的基准，用于评估在退化文档理解中OCR幻觉的问题。它包含了身份证和发票的测试样本，这些样本经过现实世界退化情况的模拟，以降低OCR的可靠性。该数据集旨在帮助评估模型在退化条件下辨别视觉信息的可靠性并作出正确反应的能力。KIE-HVQA is a dedicated benchmark designed to evaluate OCR hallucination in degraded document understanding. It encompasses test samples from identity cards and invoices, enhanced with simulated real-world degradations that compromise OCR reliability. This benchmark aims to assess models abilities to discern reliable visual information under degraded conditions and respond appropriately.

搜集汇总

背景与挑战

背景概述

KIE-HVQA是首个专门评估退化文档理解中OCR幻觉问题的基准数据集，包含身份证和发票样本，并模拟现实退化条件以测试模型在不确定数据下的表现。数据集基于CC BY 4.0许可，旨在帮助模型避免在视觉模糊情况下产生幻觉内容。