This work proposes a method for learning features from a batch of 2D sonar images to predict a multi-view point-cloud for achieving a dense 3D-reconstruction. In comparison to vision-based sensors, ...