Recently, the 3D reconstruction problem based on deep learning is one of the most important topics in computer vision. Most models use CNN-based filters to extract features utilizing local contextual information. In this paper, to improve the performance of Probability Volume, we propose to add global contextual information by Atrous Spatial Pyramid Pooling Block to the 2D Feature Extractor and 3D-ASPP Block after 3D-Regression of Cost Volume. Experimental results on DTU benchmark dataset shows performance improvement of the proposed method without increased overhead in memory usage. In addition, the proposed model shows a faster training convergence speed.