Datasets and Results for "Natural Transports-Inspired Differential Semantic Communications: Intelligent Transceivers Carry Just Differences as Meanings"

This repository contains the datasets of Differential Semantic Communications (DSCs). DSC is a closed-loop communication architecture, where intelligent parties communicate like humans transport meanings. In the DSC model, only informational differences are coded, which brings bandwidth efficiency. For benchmarking and ease of comparison by researchers on the same test image data (337 images of common objects), we have also placed the corresponding results for the compared methods of PNG, JPEG, IMAT Compressed Sensing, and Language-oriented Semantic Communications. There are two main folders with the following names: 1) DSC_Dataset: This folder contains all data related to the proposed DSCs. 2) ResultsOfOtherMethods: This folder contains the results corresponding to each compared method. Note: Each folder has a “readme” file that explains its content in more details. Reference: Please cite the following paper. [1] A. Taimori, M. Sellathurai, and T. Ratnarajah, “Natural transports-inspired differential semantic communications: Intelligent transceivers carry just differences as meanings”, IEEE Transactions on Machine Learning in Communications and Networking, 2026. Related papers to this project: There are also the conference version of DSCs [2] and the information theory of DSCs [3] as follows. [2] A. Taimori, M. Sellathurai, and T. Ratnarajah, “Differential semantic communications for salient object transmission in 6G multimedia systems”, IEEE International Conference on Computer Communications (INFOCOM’25), pp. 1-6, 2025. [3] A. Taimori, M. Sellathurai, and T. Ratnarajah, “Information-theoretic measures for closed-loop differential semantic communications”, Authorea Preprints, 2025. DOI: https://doi.org/10.36227/techrxiv.174002534.40917391/v1