Quantifying curb appeal

Visual Interpretability of Image-based Real Estate Appraisal

“…Style aspects such as the type of roof affect the aesthetics and increase the price [1]. Similar to the quality signs of the landscape, factors indicating the house's maintenance (e.g., fresh paint, no broken objects) correspond to a higher real estate value [6]. In general, the condition and first impression of the dwelling seem to correlate with the appeal and the price [15].…”

“…apartment block, street, church, yard, waterfall). The exterior image is classified into these categories and the probabilities of the images belonging to the different classes are then used as additional variables in a downstream hedonic model [6]. Alternatively, one can label an own dataset as Poursaeed et al did [23].…”

“…Setting our experiment in the context of related work, instead of using a one-stage modeling approach [3,8], we use a two-stage modeling approach like in [6]. The reasons for using a two-stage approach is that visual interpretability techniques like RAM can be applied on the CNN to increase its interpretability.…”

“…For the CNN f , we use different architectures. Model 1a and 1b are based on the VGG-16 architecture [45] following previous related research [6,7]. We customize the VGG-16 model by deleting its fully connected layers, to be able to apply RAM.…”

“…This type of neural network has the ability to spatially decompose images into edges and textures for an analysis. Both resulted in an increased predictive performance [1,3,5,6,7,8,9]. Nevertheless, the use of these algorithms led to more intransparency, limiting the interpretability of CAMA systems.…”