rects
收藏资源简介:
我们收集并构建了一个实用且具有挑战性的多方向自然场景文本数据集 (ReCTS),其中包含 25,000 张图像,其中包含大量招牌。在数据集中,所有文本行和字符都标有位置和字符代码。此外,本次比赛还提出了四个任务:(1)招牌上的端到端识别,(2)招牌上的文本行定位,(3)招牌上的字符识别,(4)招牌上的文本行识别标志牌。 我们希望该数据集和任务能够极大地促进中文文本检测和识别的研究。
We collected and constructed a practical and challenging multi-orientation natural scene text dataset (ReCTS), which consists of 25,000 images featuring a large number of signboards. Within this dataset, all text lines and individual characters are annotated with their bounding box positions and corresponding character codes. Furthermore, four tasks are proposed for this competition: (1) End-to-end recognition on signboards, (2) Text line localization on signboards, (3) Character recognition on signboards, (4) Text line recognition on signboards. We anticipate that this dataset and the proposed tasks will significantly advance research in Chinese text detection and recognition.




