Citation: Jin, J.; Xu, H.; Leng, B.
Adaptive Points Sampling for
Implicit Field Reconstruction of
Industrial Digital Twin. Sensors 2022,
22, 6630. https://doi.org/10.3390/
s22176630
Academic Editors: Zhihan Lv, Kai Xu
and Zhigeng Pan
Received: 23 July 2022
Accepted: 26 August 2022
Published: 2 September 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Adaptive Points Sampling for Implicit Field Reconstruction
of Industrial Digital Twin
Jiongchao Jin
1,2,
*, Huanqiang Xu
1
and Biao Leng
1
1
School of Computer Science and Engineering, Beihang University, Beijing 100191, China
2
Beijing GlacierAI Technology Co., Ltd., Beijing 100084, China
* Correspondence: jinjiongchao@buaa.edu.cn
Abstract:
Nowadays, the digital twin (DT) plays an important role in Industry 4.0. It aims to
model reality in the digital space for further industrial maintenance, management, and optimization.
Previously, many AI technologies have been applied in this field and provide strong tools to connect
physical and virtual spaces. However, we found that single-view 3D reconstruction (SVR) for DT
has not been thoroughly studied. SVR can generate 3D digital models of real industrial products
from just a single image. The application of SVR technology would bring convenience, cheapness,
and robustness to modeling physical objects in digital space. However, the existing SVR methods
cannot perform well in the reconstruction of details, which is indispensable and challenging in
industrial products. In this paper, we propose a new detail-aware feature extraction network based
on a feature pyramid network (FPN) for better detail reconstruction. Then, an extra network is
designed to combine convolutional feature maps from different levels. Moreover, we also propose
a novel adaptive points-sampling strategy to adaptively change the learning difficulty according
to the training status. This can accelerate the training process and improve the fine-tuned network
performance as well. Finally, we conduct comprehensive experiments on both the general objects
dataset ShapeNet and a collected industrial dataset to prove the effectiveness of our methods and the
practicability of the SVR technology for DT.
Keywords: deep learning; digital twin; implicit field; 3D reconstruction
1. Introduction
With the rapid development of the Internet of things (IoT) in Industry 4.0, the interac-
tion between physical and digital spaces has become an essential process in the industrial
world. The digital twin (DT) [
1
–
3
] was brought out to bidirectionally bridge the virtual
world and reality. DT is a complex concept, closely related to robotics technology, com-
puter vision, computer graphics, artificial intelligence, and other areas in computer science.
It enables real-time modeling, monitoring, analysis, prediction, and control of physical
objects [
4
,
5
]. DT can also significantly improve industry chain collaboration, urban man-
agement, and industrial system optimization [
6
–
9
]. Importantly, the tasks of the digital
twin cannot be accomplished without 3D models.
The 3D digital model plays an indispensable role throughout the different stages of
industrial production, including industrial product design, manufacturing, and mainte-
nance. However, the interaction between authentic products and 3D digital models is still
a challenging problem. It is also an important research topic in DT. In existing systems,
sophisticated sensors are needed to capture the 3D data of products to reconstruct 3D
digital models. In our paper, we try to utilize single-view 3D reconstruction (SVR) tech-
nology to reconstruct 3D models from one 2D image of the real world. Compared with
complex sensor-based model reconstruction, an image input can be easily obtained by
mobile devices and processed with less computing power, which enables SVR to be applied
in a wide range of scenarios.
Sensors 2022, 22, 6630. https://doi.org/10.3390/s22176630 https://www.mdpi.com/journal/sensors
