An Evaluation Method for Multiview Surface Reconstruction Algorithms

We propose a new method, 3DGiC, for evaluating the performances of various multiview surface reconstruction methods. It does not need a complete ground truth model, providing a much wider range of applications. More importantly, most existing methods only measure the quality of reconstruction in a global manner but local surface details are not involved. In contrast, 3DGiC depends on both global consistency and local accuracy of reconstruction in order to deliver a more comprehensive evaluation. The key idea is to compute a cumulative distribution of the joint probability of two local surface descriptors. We also designed experiments based on both synthetic and real data to demonstrate the advantages as well as the effectiveness of 3DGiC.