Progressive UAV Video Token Communication for Bandwidth-Limited Wireless Transmission

Efficient unmanned aerial vehicle (UAV) video transmission remains challenging under limited bandwidth and varying wireless channel conditions. Existing video transmission methods typically rely on predictive residual coding, which may suffer from limited spatiotemporal modeling capability and temporal error propagation. To address these challenges, we propose a progressive UAV video token communication framework. The proposed framework encodes video clips into compact spatiotemporal tokens, providing an efficient representation for wireless transmission. To improve robustness against channel noise, the token representations are adaptively adjusted along the temporal and spatial dimensions according to the estimated channel SNR. Moreover, progressive transmission is achieved by varying the number of active latent channels according to the target transmission rate, enabling flexible operation under different bandwidth constraints. Extensive experiments on the UAVDT dataset demonstrate that the proposed framework achieves superior reconstruction quality, perceptual quality, and semantic consistency over existing baselines under bandwidth-limited wireless conditions, while maintaining low encoding and decoding latencies of 1.86 and 3.01 ms per frame, respectively.