Special Issue on Intelligent Communications for Real-Time Computer Vision Submission Date: 2024-11-01 Real-time Computer Vision (CV) is one of the essential building blocks for a wide range of emerging applications, including digital manufacturing, future transportation, telemedicine, robotics, Virtual/Augmented/Extended/Mixed Reality (VR/AR/XR/MR), digital twin and metaverse. However, real-time performance heavily depends on communication networks, sensors and computing resources disaggregated across different devices or locations over the end-to-end communications-compute continuum. While the performance of telecommunication networks has seen significant improvements over the past decades, there is a disconnect between the Key Performance Indicators (KPIs) used to measure network performance (e.g., data rate, latency, spectral efficiency, connection density, reliability, mobility, etc.) and those used in CV, which are typically associated with specific tasks. For example, in image semantic segmentation, a set of KPIs including Average Precision (AP) and mean Intersection Over Union (mIoU) are used to measure task performance. There is therefore need for better understanding the relationship and inter-dependencies between the two kinds of KPIs in the context of real-time applications. Furthermore, the performance of existing CV algorithms under variable levels of latency, jitter, and packet loss in communication systems remains relatively unexplored. For example, existing communication systems are designed to recover the original images or videos at the receiver side. This data-oriented design principle is not suitable for machine-type computer vision applications where high-fidelity reconstructions are often not required. Task-oriented design for emerging CV applications could be a promising solution. This fundamental difference reveals the gap between bit-level transmission built on Shannon’s theory, and the requirements of completing tasks successfully and safely in automation systems. Therefore, novel design principles and methodologies such as artificial intelligence for communications and CV are in urgent need.

This Special Issue (SI) is seeking interdisciplinary and integrated contributions to tackle the communication challenges for a wide range of real-time CV applications.

Topics of interest include, but are not limited to:

Communication system designs for real-time CV applications

- Network architectures and protocols for CV applications

- Task/goal-oriented communications for CV applications

- Network programmability, intelligence, and function virtualization for CV applications

Real-time CV algorithm design in the presence of latency, jitter, & packet losses in communications

- 3D scene representations with delayed images

- Split/Federated learning for distributed CV

- Real-time CV for robotics and automation

Fundamentals and applications in communications and CV

- Joint source and channel coding and semantic communications

- Fidelity-timeliness trade-off

- Computation and compression under latency or/and energy consumption constraints

- Simultaneity, causality, and reasoning in real-time interactions

- Real-time communications and CV for Metaverse

- Recent results from testbeds, prototypes, and experiments/trails