PaHCC
收藏资源简介:
PaHCC数据集由中国科学院自动化研究所创建,专注于手写汉字识别的领域泛化研究。该数据集包含280种打印字体和720种手写字体的996478个样本,旨在通过合成打印数据训练模型,使其能够直接泛化到扫描手写数据的识别。数据集的创建过程涉及从SCUT-SPCCI数据库解码合成打印字符图像,以及从CASIA-HWDB1.0-1.1解码扫描手写页面。PaHCC数据集的应用领域主要集中在提高模型对未见领域数据的泛化能力,特别是在手写汉字识别任务中,解决数据分布偏移问题。
The PaHCC dataset, created by the Institute of Automation, Chinese Academy of Sciences, focuses on domain generalization research for handwritten Chinese character recognition. This dataset contains 996,478 samples spanning 280 printed font types and 720 handwritten font types, aiming to train models via synthetic printed data so that they can directly generalize to the recognition of scanned handwritten data. The dataset creation process involves decoding synthetic printed character images from the SCUT-SPCCI database and decoding scanned handwritten page documents from the CASIA-HWDB1.0-1.1 database. The application scenarios of the PaHCC dataset mainly focus on improving the generalization ability of models to unseen domain data, especially in handwritten Chinese character recognition tasks, to address the data distribution shift problem.

- 1Towards Reliable Domain Generalization: A New Dataset and Evaluations中国科学院自动化研究所 · 2023年



