A Multi-resolution Adaptive Encoding Method for 3-D Reconstruction of DPO-SDF
Yanpeng Jia, Yuxuan Li, Ce Ding, Lizhen Yang, Hai Lin · 2024
Recently, the differentiable physical optics method (DPO-SDF) achieves high-precision 3-D reconstruction of perfect electric conductor (PEC) scatterers from scattered electric fields. Nevertheless, employing implicit neural networks directly as the 3-D representation of the reconstructed target results in significant memory consumption and high computational costs during training stage. This paper presents a multi-resolution adaptive encoding method to significantly reduce the number of parameters that require updating in each iteration. Numerical experiments demonstrate that our encoding method significantly reduces memory usage during the training stage while improving reconstruction speed without compromising accuracy.