未明确提及

本研究涉及的数据集用于训练自动驾驶中的语义分割模型，主要包含合成图像和来自Carolo-Cup环境的实际图像。数据集的创建结合了模拟生成的合成图像与真实环境中的图像，采用监督学习方法，每张图像都有相应的地面实况。该数据集旨在减少标注工作量，通过模拟生成路线，无需实际复制。此外，数据集还用于评估不同视角（如第一人称视角和鸟瞰视角）对模型性能的影响，以及不同计算精度（如16位和32位浮点运算）对实时性能和准确性的影响。

The dataset involved in this study is used for training semantic segmentation models in autonomous driving, mainly including synthetic images and real-world images collected from the Carolo-Cup environment. This dataset is constructed by combining simulation-generated synthetic images and real-world captured images, adopting a supervised learning approach, where each image has its corresponding ground truth. The dataset aims to reduce annotation workload by generating driving routes via simulation without the need for physical replication. Additionally, this dataset is also utilized to evaluate the impact of different perspectives (e.g., first-person perspective and bird's-eye view) on model performance, as well as the effects of varying computational precisions (e.g., 16-bit and 32-bit floating-point operations) on real-time performance and prediction accuracy.