SliceVision-F2I: Synthetic Visual Dataset for Network Slicing
Description
SliceVision-F2I is a new dataset created to change network slice Key Performance Indicators (KPIs) into different visual formats, allowing the use of computer vision methods in managing network slicing. The dataset contains 30,000 samples per generation method, each corresponding to one of three key network slice types: Enhanced Mobile Broadband (eMBB), Ultra-Reliable Low-Latency Communications (URLLC), and Massive Internet of Things (mIoT). Each sample in the dataset is transformed into four distinct visual representation patterns, namely Physically-Guided Patterns, Perlin Noise Patterns, Wallpaper Patterns, and Fractal Branching Patterns. These representations are designed to capture different structural and spatial characteristics of KPI data, allowing for diverse visual interpretations and model generalization. SliceVision-F2I supports a wide range of use cases. In network management, it can help spot problems by looking for unusual patterns, categorise different types of network slices based on what it sees, and estimate user experience quality using statistical models. For machine learning research, the dataset facilitates multimodal learning by offering multiple visual perspectives of the same data. It also enables experimentation with data augmentation techniques across different representation styles and supports the development of explainable AI methods by offering interpretable visual formats. In educational settings, SliceVision-F2I serves as a valuable resource for teaching network performance concepts through visual demonstration. It provides a standardized benchmark for evaluating AI-based approaches in telecommunications and is well-suited for rapid prototyping of visual analytics tools. The dataset, along with sample usability code, is publicly available at: https://github.com/AbidHasanRafi/SliceVision-F2I
Files
Institutions
- Hajee Mohammad Danesh Science and Technology University