Constructing the Depth Map of a Microdevice
Keywords
Loading...
Authors
Issue Date
2022-07-21
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Over the past decade, neural networks have been able to perform increasingly well on the
ill-posed problem where a 3D representation is reconstructed from 2D stereo data. Nowadays
even unsupervised neural networks show decent performance on this task, most notably with
an auto-encoder architecture. These networks are often tested on large public data sets of
real-world data (like KITTI). We applied this auto-encoder architecture to a data set of
image pairs of semiconductors. Due to the small size of the data set, we segmented the
image pairs into smaller patches that will be used as training data. Several experiments
have been conducted to test the optimal segmentation of image pairs. These experiments
showed mixed results on the ability to create a depth map from the input images, where the
main di culty appeared to be distinguishing the wires from the background surface.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen