Abstract: Multi-view based reconstruction is one of the central problems in computer vision. In recent years, many excellent algorithms have been reported, such as the PMVS by Furukawa. However, we observed that the following two aspects of the PMVS need further improvements. The first one is that the obtained normal of reconstructed point is not well consistent with its local geometry, and the problem becomes more pronounced under certain image capturing configuration such as downward-shooting or upward-shooting, a frequent practice in large scene reconstruction. The second is its inherent space and time complexity. Especially with high resolution images, its space and time loads become unaffordable. In this work, two remedies are proposed. We propose a patch adjusting trick through the scene geometric information to enhance the patch's normal estimation, and a multi-resolution expanding tactic to well balance the computational cost and the reconstruction accuracy. The experiments demonstrate the effectiveness and practicability of our improved algorithm.