一篇新论文修订了 RVL-CDIP 数据集，这是一个流行的文档分类器基准。该论文通过识别和纠正标签错误及测试-训练重叠问题来完成修订。分析显示，原始数据集中约有 12% 的标签错误和 35% 的测试-训练重复。对数据集的修改表明，移除标签错误提高了分类准确性，而移除重复项则导致准确性下降。此外，在修正后的数据上进行训练，显著增强了分布外泛化能力，监督模型在 RVL-CDIP-N 基准上的平均准确率提高了 8.1 个百分点。 AI
影响 文档分类基准数据集质量的提高可能带来更可靠的模型评估和更好的分布外泛化能力。
排序理由 该集群包含一篇详细介绍数据集修订和基准分析的学术论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2606.31446v1 Announce Type: new Abstract: RVL-CDIP is a popular dataset for benchmarking document classifiers. However, the dataset contains ample amounts of label errors as well as non-trivial amounts of test-train overlap, both of which may impact model performance metric…
RVL-CDIP is a popular dataset for benchmarking document classifiers. However, the dataset contains ample amounts of label errors as well as non-trivial amounts of test-train overlap, both of which may impact model performance metrics. In this paper, we address these two problems …