Paper Proposes Channel-Adaptive Graph Carriers for Semantic Image Communication

A new arXiv preprint introduces a method for semantic image communication that uses channel-adaptive region adjacency graphs as carriers instead of dense latent tensors or grid-aligned semantic layouts. The approach aims to explicitly encode relationships between image regions so that task-relevant scene structure survives tight channel budgets. The abstract frames the work as addressing a gap in how region-level relations are represented under limited bandwidth.