RnD

RnD数据集是由卡内基梅隆大学DEVCOM陆军研究实验室创建，专注于越野摩托车赛车号码的光学字符识别研究。该数据集包含2,411张由专业摄影师拍摄的图像，展示了摩托车赛车手在越野比赛中的情况。数据集中的图像因泥浆遮挡、运动模糊、非标准字体、反光、复杂背景等因素而使OCR变得困难。数据集共有5,578个手动标注的围绕可见摩托车号码的边界框，以及转录的数字和字母。RnD数据集旨在推动在不受约束的自然环境中稳健识别文本的能力，适用于赛车分析、计时系统、媒体广播等领域。

The RnD Dataset was developed by the U.S. Army DEVCOM Army Research Laboratory at Carnegie Mellon University, focusing on optical character recognition (OCR) research for off-road motorcycle race numbers. This dataset includes 2,411 images captured by professional photographers, showcasing motorcycle racers during off-road competitions. The images in this dataset present challenges for OCR due to factors such as mud occlusion, motion blur, non-standard fonts, reflections, and complex backgrounds. It contains a total of 5,578 manually annotated bounding boxes surrounding visible motorcycle race numbers, alongside transcribed digits and letters. The RnD Dataset is intended to promote robust text recognition capabilities in unconstrained natural environments, and is applicable to fields such as race analysis, timing systems, and media broadcasting.