Pattern Matters: The Impact of Projection Patterns on Deep Learning 3-D Profilometry

In the 3-D optical metrology, single-shot deep learning-based structured light profilometry (SS-DL-SLP) has gained attention because of its measurement speed, simplicity of optical setup, and robustness to noise and motion artifacts. The chosen projection pattern is a key feature within structured light profilometry (SLP), with or without a neural network. Algorithmic processes using multishot or single-shot instance scanning require in-depth analyses to determine the optimal pattern for their intended measurement purposes. This article determines whether the transition to deep learning still requires such in-depth research to find the most optimal projection pattern. Our study highlights the performance of networks when unconventional projection patterns are used and analyzes what hyperparameters, combined with the pattern analyses, make models perform better. We ensure that these performance optimizations are not model dependent by executing the same training task on different model structures. All code is publicly available to facilitate further research and promote reproducibility: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/RhysEvan/Pattern-Matters</uri>