Reproduce “Land-then-transport: A Flow Matching-Based Generative Decoder for Wireless Image Transmission” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jingwen Fu, Ming Xiao, Mikael Skoglund, Dong In Kim
The Land-then-Transport (LTT) generative decoder employs Flow Matching to embed wireless channel characteristics into a continuous-time probability flow, enabling a low-latency, channel-aware approach for image transmission. It achieves higher perceptual quality and fidelity than prior methods while requiring significantly fewer decoding steps.
Key insights
- Flow Matching (FM) allows for the learning of a deterministic continuous-time velocity field, enabling significantly faster inference compared to the stochastic iterative processes of diffusion models.
- Integrating the physical wireless channel's noise characteristics directly into the continuous probability flow creates a channel-aware generative decoder, making the reconstruction process inherently adaptive.
- Complex wireless channels (Rayleigh fading, MIMO) can be reduced to an AWGN-equivalent form via linear preprocessing, allowing a single AWGN-trained generative decoder to be reused universally without retraining.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.