Source Code for Structure-Preserving Semantic Image Transmission under Low-Bandwidth Networks

This dataset provides the implementation code for a structure-preserving semantic transmission method for low-bandwidth networks, which is designed for image semantic transmission, structure-preserving reconstruction, and related experimental validation under bandwidth-constrained conditions. The proposed method targets ultra-low bitrate visual communication scenarios and addresses the “digital cliff” effect of traditional coding schemes as well as the difficulty of existing semantic transmission approaches in balancing structural fidelity, color consistency, and low latency. A multi-stream decoupled image transmission framework consisting of semantic, structural, and appearance representations is proposed.The code mainly includes the following components: (1) a semantic gateway with object-aware semantic rate-distortion optimization; (2) a structure stream encoding module based on edge extraction; (3) a low-frequency appearance stream extraction and compression module; (4) multimodal feature decoupling implementation; (5) a deterministic dual-stream generative decoding network based on SPADE; (6) boundary consistency processing and elastic gradient alignment for block-wise transmission; and (7) experimental and evaluation scripts. These modules support the proposed structure-preserving semantic transmission model and its experimental pipeline.The method is evaluated on the Set14 and DIV2K datasets, and its performance is further analyzed under LoRa physical link conditions in terms of transmission efficiency, latency, and reconstruction quality in low-bandwidth environments. Experimental results show that under a low-bitrate constraint of approximately 0.3 bpp, the proposed method achieves a good balance among structural consistency, visual naturalness, and end-to-end inference latency, making it suitable for high-reliability edge visual communication scenarios.This dataset can be used for research and experimental reproduction in areas such as semantic communication, low-bitrate image transmission, generative reconstruction, edge intelligence, and high-reliability visual communication. Detailed instructions, runtime environment, and dependency configurations can be found in the accompanying README file.