Abstract
Existing wireless video transmission schemes directly conduct video coding inpixel level, while neglecting the inner semantics contained in videos. In thispaper, we propose a wireless video semantic communication framework,abbreviated as WVSC, which integrates the idea of semantic communication intowireless video transmission scenarios. WVSC first encodes original video framesas semantic frames and then conducts video coding based on such compactrepresentations, enabling the video coding in semantic level rather than pixellevel. Moreover, to further reduce the communication overhead, a referencesemantic frame is introduced to substitute motion vectors of each frame incommon video coding methods. At the receiver, multi-frame compensation (MFC) isproposed to produce compensated current semantic frame with a multi-framefusion attention module. With both the reference frame transmission and MFC,the bandwidth efficiency improves with satisfying video transmissionperformance. Experimental results verify the performance gain of WVSC overother DL-based methods e.g. DVSC about 1 dB and traditional schemes about 2 dBin terms of PSNR.