Share Email Print
cover

Proceedings Paper • new

Classification of real estate images using transfer learning
Author(s): Yang Cao; Shinichi Nunoya; Yusuke Suzuki ; Masachika Suzuki; Yoshio Asada; Hiroki Takahashi
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

With the growing mobility of the population and popularity of the Internet, real estate agents have larger database to manage. This paper presents a solution to classify images of a certain house, such as living room, kitchen, bathroom, layout sketch and external appearance collected by a real estate agent using transfer learning. The pictures are like those images posted on the real estate agent website to help people find out what’s the house looks like inside and outside. We employ a transfer learning approach for VGG-19 architecture. Using a network pre-trained on the general ImageNet dataset, we perform supervised fine-tuning on the last full connect layer and change the output size from 1000 to 5. Experimental results achieved with 5-fold cross-validation show that after training, this fine-tuning approach achieves high test accuracy of 99.4%.

Paper Details

Date Published: 6 May 2019
PDF: 6 pages
Proc. SPIE 11069, Tenth International Conference on Graphics and Image Processing (ICGIP 2018), 110691I (6 May 2019); doi: 10.1117/12.2524417
Show Author Affiliations
Yang Cao, The Univ. of Electro-Communications (Japan)
Shinichi Nunoya, AVANT Corp. (Japan)
Yusuke Suzuki , AVANT Corp. (Japan)
Masachika Suzuki, AVANT Corp. (Japan)
Yoshio Asada, AVANT Corp. (Japan)
Hiroki Takahashi, The Univ. of Electro-Communications (Japan)


Published in SPIE Proceedings Vol. 11069:
Tenth International Conference on Graphics and Image Processing (ICGIP 2018)
Chunming Li; Hui Yu; Zhigeng Pan; Yifei Pu, Editor(s)

© SPIE. Terms of Use
Back to Top