On Splitting Lightweight Semantic Image Segmentation for Wireless Communications

2025-07-14

AI Summary

To address the challenges of high computational overhead, excessive transmission bandwidth, and compromised segmentation accuracy in semantic image segmentation for resource-constrained and dynamically varying channel environments—such as those envisioned for 6G—this paper proposes an edge–end collaborative segmentation framework. The framework intelligently partitions a lightweight semantic segmentation model across the transmitter (front-end feature extraction) and receiver (back-end decoding and segmentation), enabling joint optimization of computation and communication loads. It integrates adaptive model partitioning strategies, compact network architecture design, and a multi-dimensional evaluation mechanism jointly considering bit-rate, computation, and accuracy. Experimental results demonstrate that, compared to end-to-end transmission, the proposed approach reduces transmission bit-rate by up to 72%, decreases transmitter-side computational load by over 19%, and maintains segmentation accuracy nearly unchanged—achieving a significant trade-off between bandwidth efficiency and edge deployment feasibility.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Learning on the Edge & Model Compression
• Search and Optimization: Distributed Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Cloud, edge and content delivery systems for the Web
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Semantic communication represents a promising technique towards reducing communication costs, especially when dealing with image segmentation, but it still lacks a balance between computational efficiency and bandwidth requirements while maintaining high image segmentation accuracy, particularly in resource-limited environments and changing channel conditions. On the other hand, the more complex and larger semantic image segmentation models become, the more stressed the devices are when processing data. This paper proposes a novel approach to implementing semantic communication based on splitting the semantic image segmentation process between a resource constrained transmitter and the receiver. This allows saving bandwidth by reducing the transmitted data while maintaining the accuracy of the semantic image segmentation. Additionally, it reduces the computational requirements at the resource constrained transmitter compared to doing all the semantic image segmentation in the transmitter. The proposed approach is evaluated by means of simulation-based experiments in terms of different metrics such as computational resource usage, required bit rate and segmentation accuracy. The results when comparing the proposal with the full semantic image segmentation in the transmitter show that up to 72% of the bit rate was reduced in the transmission process. In addition, the computational load of the transmitter is reduced by more than 19%. This reflects the interest of this technique for its application in communication systems, particularly in the upcoming 6G systems.

Problem

• Balancing computational efficiency and bandwidth in semantic image segmentation
• Reducing transmitter load in resource-limited wireless communication environments
• Maintaining segmentation accuracy while minimizing transmitted data and bit rate

Innovation

• Splits semantic segmentation between transmitter and receiver
• Reduces bandwidth by minimizing transmitted data
• Lowers transmitter computational load significantly