Most of the existing learning-based multiview stereo (MVS) approaches reconstruct depth map through supervised models that are trained with large amounts of data without any prior knowledge about the ...