HDR-Net: A HDR dataset of fringe projection based on deep network

In recent years, network-based fringe-projection-profilometry (FPP) has achieved significant advancements in reconstructing high-dynamic-range (HDR) surfaces. However, a common dataset and benchmark for evaluating the performance of various algorithms are still lacked that obviously obstacle this field’s development. A FPP dataset that totally contains 720 HDR scenes was collected in this paper. In addition, a state-of-art FPP network achieving the effect of 6-step phase-shift (PS) algorithm with only 3 images was proposed. Compared with the traditional networks, it is a two-stage network: it firstly outputs the texture of the object surface, and secondly takes the texture and the original images as inputs to the network, predicting the numerator and denominator of the wrapped phase φ as outputs. Experiments have proved the validness: the phase error of this method is 2.61×10^(-4) (2π∙rad), whereas the phase error of 6-step PS algorithm is 2.59×10^(-4) (2π∙rad). Dataset and code are available at the website: https://github.com/SHU-FLYMAN/HDR-NET.