A Semi-Supervised Method for PatchMatch Multi-View Stereo with Sparse Points

Recently, the deep-learning-based PatchMatch method has been rapidly developed in 3D reconstruction, based on which boundary regions are filled with other parts that most closely match edge parts, but limited data hinder generalization of to unknown settings. If various large-scale datasets generated, process would require considerable time and resources when performing neighborhood point-matching calculations using random iterative algorithms. To solve this issue, we first propose a new, sparse, semi-supervised stereo-matching framework called SGT-PatchMatchNet, can reconstruct reliable structures small number points ground truth surface frame values. Secondly, order problem luminosity inconsistency some pixels views, photometric similar-point loss function is proposed improve performance causes information project depth value predicted meet same coordinates. Finally, blurring map obtained network model, robust-point consistency integrity robustness occlusion areas. The experimental results show not only good visual effects indicators also effectively reduce amount computation calculation time.