Matching cost aggregation plays a fundamental role in learning-based multi-view stereo
networks. However directly aggregating adjacent costs can lead to suboptimal results due to …