As analysis of imagery and other science data plays a greater role in mission execution, there is an increasing need for autonomous marine vehicles to transmit these data to the surface. Communicating imagery and full-resolution sensor readings to surface observers remains a significant challenge. Yet, without access to the data acquired by an unmanned underwater vehicle (UUV), surface operators cannot fully understand the mission state of a vehicle. This paper presents an architecture capable of multihop communication across a network of underwater acoustic relays. In concert with an abstracted physical layer, CAPTURE provides an end-to-end networking solution for communicating science data from autonomous marine vehicles. Automatically selected imagery, SONAR, and time-series sensor data are progressively transmitted across multiple hops to surface operators. To incorporate human feedback, data are transmitted as a sequence of gradually improving data “previews.” Operators can request arbitrarily high-quality refinement of any resource, up to an error-free reconstruction. The results of three diverse field trials on SeaBED, OceanServer, and Bluefin AUVs, with drastically different software architectures, are also presented.