HUST-ART, HUST-AST, ABE, Tana

本研究介绍了四个针对阿姆哈拉语场景文本检测和识别的综合基准数据集：HUST-ART、HUST-AST、ABE和Tana。这些数据集由华中科技大学创建，旨在解决阿姆哈拉语文本识别的挑战，如视觉复杂性、图像质量差和文本间断性。HUST-ART包含2200张自然场景图像，用于训练和测试，涵盖多种场景。HUST-AST则包含75,904张图像，用于合成文本实例。ABE和Tana分别包含真实和合成文本图像，用于文本识别任务。这些数据集的应用领域包括提升阿姆哈拉语文本检测和识别算法的性能，解决特定于阿姆哈拉语的文本识别问题。

This study introduces four comprehensive benchmark datasets for Amharic scene text detection and recognition: HUST-ART, HUST-AST, ABE, and Tana. Developed by Huazhong University of Science and Technology (HUST), these datasets are designed to tackle core challenges in Amharic text recognition, such as visual complexity, poor image quality and text discontinuity. HUST-ART contains 2,200 natural scene images for both training and testing, covering diverse scenarios. HUST-AST consists of 75,904 synthetic text instance images. ABE and Tana respectively provide real and synthetic text image datasets for text recognition tasks. The applications of these datasets focus on improving the performance of Amharic text detection and recognition algorithms and solving Amharic-specific text recognition issues.