Abstract
Video semantic communication (SemCom) focuses on transmitting the semantic meaning of the video rather than pixel-accurate frames, enabling efficient operation under bandwidth constraints. This paper presents a trajectory-centric semantic video communication framework that performs predictive video transmission in a compact latent space using a small set of context frames. The proposed model leverages vector-quantized latent compression with a lightweight temporal transformer to infer future frames at the receiver, significantly reducing transmission overhead. Motion dynamics are explicitly treated as semantic information, providing a compact representation of motion fidelity. Using normalized end-point error (nEPE), a novel trajectory-based matrix for video SemCom, we measured motion accuracy. Experimental results on the BAIR robot pushing dataset demonstrate that the proposed approach preserves motion trajectories and perceptual quality while operating at a low bitrate of 0.27 bits per pixel (bpp), outperforming conventional video codecs and learning-based baselines in semantic fidelity under wireless channel conditions.
| Original language | English |
|---|---|
| Title of host publication | 2026 IEEE International Mediterranean Conference on Communications and Networking, MeditCom 2026 |
| Place of Publication | Piscataway, NJ |
| Publisher | IEEE |
| ISBN (Electronic) | 9798331551438 |
| ISBN (Print) | 9798331551445 |
| DOIs | |
| Publication status | Published - 13 Aug 2026 |
| Event | IEEE International Mediterranean Conference on Communications and Networking 2026 - Cagliari, Italy Duration: 6 Jul 2026 → 9 Jul 2026 https://meditcom2026.ieee-meditcom.org/ (Link to conference website) |
Publication series
| Name | IEEE International Mediterranean Conference on Communications and Networking (MeditCom) |
|---|
Conference
| Conference | IEEE International Mediterranean Conference on Communications and Networking 2026 |
|---|---|
| Abbreviated title | IEEE MeditCom 2026 |
| Country/Territory | Italy |
| City | Cagliari |
| Period | 6/07/26 → 9/07/26 |
| Internet address |
|
Funding
This work was supported by the Research Council of Finland through the 6G Flagship (grant no. 369116).
Keywords
- generative AI
- semantic communication
- semantic video communication
- trajectory modeling
- transformer
- world model
Fingerprint
Dive into the research topics of 'Trajectory-centric generative semantic communication for wireless video transmission'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver