Abstract
Semantic Communication (SemCom) has garnered widespread attention due to its effectiveness and intelligence as an emerging technique. The current image SemCom system, based on Deep Joint Source-Channel Coding (JSCC), requires joint model training at both ends, leading to coupled models that need concurrent updates. Additionally, the semantics transmitted via joint training are feature vectors, which lack interpretability. To address these issues, we design an Explainable Semantics-based Image Semantic Communication (ES-ISC) demo. This demo transforms images into explainable semantic texts and segmentation maps for transmission. By leveraging the universality of these semantic carriers, we facilitate the decoupling of transmitter and receiver training without substantially impacting performance. Moreover, the interpretability of the designed semantic carriers supports multiple downstream tasks. Experimental results demonstrate that our system can transmit images with over 100fold compression while maintaining high-quality reconstruction at the receiver.