In this paper, we proposed an interior design system.
We have implemented the scene selection function and the
house type drawing function to get the apartment type. After
getting the basic apartment type, we have also provided some
other basic decoration functions such as furniture placement,
furniture conversion, material conversion, and light switch. These
functions are operated by mouse clicking and keyboard control.
In addition, we have added some AI modules to provide an
additional assistant. Through the recognition of the picture, the
texture can be trained, and the ideal texture has been obtained.
The implementing environment of our design system is UE4, and
the AI algorithm was written in Python and tensor flow.
http://www.aleeshainstitute.com/interior-designing-course.php
The evolution of virtual reality (VR) technologies has been
used in numerous fields of social life. And it promotes
the traditional architectural design and interior decoration
design [1]–[7]. This new design form can display the design
more directly in front of people to meet people’s requirements
for decoration. It promotes communication between customers
and designers and enables the design to better meet the needs
of customers. This article mainly introduces the interior design
system that includes furniture placement, real-time conversion
of materials, and also explains the virtual reality technology in
the system. UE4 (unreal engine 4) is a game engine developed
by epic games. Using the UE4 engine, we developed the
functions required for the interior design described in our
system. About AI [8]–[11], we used CycleGAN [12]. Since
the texture making process has always been a trouble to
new beginners. In this paper, we have tried to implement
CycleGAN in UE4 to transform texture style by means of
CycleGAN’s strong ability of style translation. Basically, the
principle of Generative Adversarial Networks (GANs) is to
create an adversarial loss that is used for the differentiation
of produced images and real images. But all existing works
require paired training examples, which refer to the concept
of image-to-image translation. The main work of CycleGAN
This work was supported in part by the National Natural Science Foundation
of China under Grant 61872241 and Grant 61572316, in part by the Macau
Science and Technology Development Fund under Grant 0027/2018/A1, in
part by the National Key Research and Development Program of China under
Grant 2017YFE0104000 and Grant 2016YFC1300302, and in part by the
Science and Technology Commission of Shanghai Municipality under Grant
18410750700, 17411952600, and Grant 16DZ0501100.
is capture special features of an image group and assuming
how these features can be interpreted into alternative image
group, not in the presence of matching training examples
which is also its biggest advantage. These are the following
main contributions for our work:
• Incorporate AI method: Incorporated AI methods to
achieve certain effects. The specific content is described
later in this article.
• Realize the choice and autonomous drawing of house
type: In the system of this article, we have implemented
two methods of house type design which selects the
existing house type and drawing a new one.
Intelligent design CycleGAN The idea of image-to-image
conversion is to pay consistence model for one input and
output training image pair. These type of ideas have applied
in different errands for producing photographs from sketches,
features, or semantic designs also these type of methods
need pairing. As a result of the short comings of imageto-image translation, different methods used to tackle the
unpaired image-to-image translation. Like [13] use crossmodel scene networks for weight-sharing problems to acquire
a mutual illustration diagonally. Another line of simultaneous
work inspires the input-output distribute the convinced content
features if they are different in styles. The basic concept relates
the idea of without pairing image-to-image translations are cycle consistency, that use transitivity as a technique to normalize
organized data. Recently, higher order cycle consistency has
applied by many types of research. In [14], [15], practices
sequence reliability loss for consuming transitivity to managed
CNN training. [16], independently uses a same purpose for
unpaired image-to-image translation.
-About interior decoration The combination of VR and
design is very significant and a lot of relevant work have
already existed. Because VR technology is very important to
Computer Integrated Engineer System, the VRML-Java based
on VR technology is used to develop the visual design system
[17]. Different people design interior decoration system by
the 3D scene modeling technique [18]. When discussing the
application of VR in interior decoration, combined with actual
development experience, related techniques such as view scene modeling, real-time rendering, roaming and sound effects under the virtual environment are described [19]. Thus, our work
has been integrated on the premise of combining the abovementioned content, and incorporated AI technology, adding
certain challenges while applying traditional technologies.
Comments
Post a Comment