Temporal Context Based Video Semantic Transmission

With the increasing demand for low-latency wireless video transmission applications, the limitations of classical separation-based coding schemes are difficult to effectively capture long-term dependencies and spatiotemporal correlations. To address this challenges, this paper proposes a novel approach called Temporal Context Based Video Semantic Transmission (TCVST) that effectively preserves motion and texture details across multiple scales. TCVST enhances both spatial and temporal accuracy, enabling the model to better captures non-uniform motion and texture variations. Experimental results show that our TCVST achieves better coding gain and Rate-distortion (RD) performance in various established metrics such as Peak Signal-to-Noise Ratio (PSNR) and Multiscale-Structure Similarity (MSSSIM). In terms of transmission performance under complex scene video datasets, the proposed TCVST method can save up to $44.4 \%$ of the channel bandwidth cost, compared to the classical H.264/H. 265 combined with low-density parity-check (LDPC) and digital modulation schemes.