Title

Aesthetic Quality Assessment for Real Estate Images through Deep Learning Methods

Abstract

Abstract

In this thesis, we aim to find the aesthetic quality of real estate images. Although aesthetic assessment is a subjective terminology, it is highly correlated with photographic rules. The aesthetic quality of images in real estate affects the decision of potential people of interest. The aesthetic evaluation of images is established via the Aesthetic Visual Assessment (AVA) dataset benchmark. Although AVA is a publicly available and diverse image dataset, it is cannot be adapted to the real estate domain. Therefore, we constructed the Real-Estate Aesthetics Assessment Dataset (RAAD), which consists of real and synthetic real estate images. In order to gather subjective user data on the RAAD, a user study is conducted on a custom web-based scoring platform, serving RAAD image data. We proposed a methodology, which benefits the state-of-the-art deep image classification models. Such models take an image as input and output the aesthetic quality score as aesthetically pleasing or unpleasing. The results are evaluated on the AVA dataset and later on RAAD.

Supervisor(s)

Supervisor(s)

NAZLI OZGE UCAN

Date and Location

Date and Location

2022-09-02 14:00:00

Category

Category

MSc_Thesis