3D Mesh Reconstruction from 2D Images: A NeRF based Approach

摘要

3D object reconstruction is a vital obstacle within computer vision, and several techniques have been proposed to tackle it. However, the automation of the reconstruction process continues to pose a significant challenge, and limited research has been devoted to this problem. We proposed a NeRF-based pipeline for robust multi-view object reconstruction. We conducted an extensive analysis of existing NeRF-based methods and addressed some of those limitations. Our proposed pipeline achieves 17.6% higher quality reconstructions at one-tenth of the time. We evaluated our pipeline using benchmark datasets, and our results show that it outperforms state-of-the-art approaches with respect to quantitative and qualitative evaluations, respectively. Our proposed pipeline offers a robust and efficient solution to the multi-view object reconstruction problem, with potential applications in several domains, including robotics, virtual, and augmented reality.