SCO dataset : A dataset for semantic communication

The integration of communication and artificial intelligence has become a development trend, one of the applications is semantic communication, but the current research lacks the support of comprehensive datasets. To solve this problem, we built a new image and video dataset, named SCO dataset, for the researches on semantic communication and computing. First, we introduce the peculiarities of the dataset, which contains 5100 images and 138 video clips. Secondly, we we give the data generation and processing methods of the dataset, including images and videos. Then we describes the labeling method of the dataset for intelligent tasks, including classfication and object detection tasks, and the evaluation method of the dataset for subjective experience of human users. Next, we give and analysize the experimental results of the dataset. Finally, we give some practical applications of this dataset, including end-to-end semantic communication, joint source and channels semantic coding, and multi-user semantic communication resource management.

通信与人工智能的融合已成为一项发展趋势，其中一项应用便是语义通信，然而，当前的研究尚缺乏全面数据集的支持。为解决此问题，我们构建了一个新的图像和视频数据集，命名为SCO数据集，以供语义通信与计算研究之用。首先，我们介绍了该数据集的独特之处，该数据集包含5100幅图像和138个视频片段。其次，我们阐述了数据集的数据生成与处理方法，包括图像和视频的处理。随后，我们描述了数据集用于智能任务的标注方法，包括分类和目标检测任务，以及用于评估人类用户主观体验的数据集评价方法。接着，我们分析了数据集的实验结果。最后，我们探讨了该数据集的一些实际应用，包括端到端的语义通信、联合源与信道语义编码以及多用户语义通信资源管理。