Both structural and contextual information is essential and widely used in image analysis. However, current multi-view stereo (MVS) approaches usually use a single common pre-trained model as pixel ...