Real-Time Indoor Scene Layout Estimation Based on Improved Lightweight Network

Youjun Yue; Jie Zhang; Hui Zhao; Hongjun Wang

doi:10.3788/LOP202259.1810007

Laser & Optoelectronics Progress, Volume. 59, Issue 18, 1810007(2022)

Real-Time Indoor Scene Layout Estimation Based on Improved Lightweight Network

Youjun Yue¹, Jie Zhang^1、*, Hui Zhao^1,2, and Hongjun Wang¹

¹Tianjin Key Laboratory of Control Theory & Applications in Complicated Systems, School of Electrical and Electronic Engineering, Tianjin University of Technology, Tianjin 300384, China

²College of Engineering and Technology, Tianjin Agricultural University, Tianjin 300392, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(9)

Fig. 1. Structure of the proposed network

Download full size

Fig. 2. Applications of different parameters in ESA network

Download full size

Fig. 3. Experimental results on LSUN dataset. (a) Original picture; (b) real layout; (c) layout obtained by proposed method

Download full size

Fig. 4. Experimental results on Hedau dataset. (a) Original picture; (b) real layout; (c) layout obtained by proposed method

Download full size

Fig. 5. Results of ablation experiment. (a) Original image; (b) real layout; (c) ablating layered monitoring (ALM); (d) ablating simple combination (ASC); (e) ablating semantic transfer (AST); (f) combination of three mechanisms (CTM)

Download full size

Table 1. Performance evaluation of different methods on the LSUN dataset

View table

Table 1. Performance evaluation of different methods on the LSUN dataset

Method	PE /%	End-to-end	Network backbone
Method in Ref.［5］	24.23	No	No
Method in Ref.［9］	16.71	No	FCN（VGG-16）
Method in Ref.［11］	10.63	No	FCN（VGG-16）
Method in Ref.［12］	9.31	No	FCN（VGG-16）
Method in Ref.［16］	9.86	Yes	Encode and decode
Method in Ref.［13］	12.49	No	Encode and decode
Method in Ref.［19］ Method in Ref.［21］	7.79 9.05	Yes Yes	SegNet×6 Encode and decode
Proposed method	7.35	Yes	ESA（ResNet-34）

Table 2. Performance evaluation of the different methods on the Hedau dataset

View table

Table 2. Performance evaluation of the different methods on the Hedau dataset

Method	PE /%
Method in Ref.［5］	21.20
Method in Ref.［7］	20.10
Method in Ref.［8］	12.80
Method in Ref.［9］	12.83
Method in Ref.［11］	9.73
Method in Ref.［12］	8.67
Method in Ref.［16］	8.36
Method in Ref.［13］	12.70
Method in Ref.［19］	7.44
Proposed method	8.32

Table 3. Results of the ablation study
View table
Table 3. Results of the ablation study
Method PE /%
ALM 7.84
ASC 7.67
AST 14.30
CTM（ours） 7.35

Table 4. Results of different loss weights
View table
Table 4. Results of different loss weights
a d₁ d₂ d₃ PE /%
0.25 0.25 0.25 0.25 8.81
0.5 0.5 0.5 0.5 7.99
0.7 0.7 0.7 0.7 8.07
1 1 1 1 7.72
1.25 1.25 1.25 1.25 8.07
1 0.5 0.5 0.5 7.67
1 0.2 0.2 0.2 7.35
1 0 0 0 7.84

Tools

Get Citation

Copy Citation Text

Youjun Yue, Jie Zhang, Hui Zhao, Hongjun Wang. Real-Time Indoor Scene Layout Estimation Based on Improved Lightweight Network[J]. Laser & Optoelectronics Progress, 2022, 59(18): 1810007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites