Deep Learning Model for Predicting Preference of Space by Estimating the Depth Information of Space using Omnidirectional Images

Deep Learning Model for Predicting Preference of Space by Estimating the Depth Information of Space using Omnidirectional Images

Kinugawa, Hina; Takizawa, Atsushi

Article:

In this study, we developed a method for generating omnidirectional depth imagesfrom corresponding omnidirectional RGB images of streetscapes by learningeach pair of omnidirectional RGB and depth images created by computergraphics using pix2pix. Then, the models trained with different series of imagesshot under different site and weather conditions were applied to Google streetview images to generate depth images. The validity of the generated depth imageswas then evaluated visually. In addition, we conducted experiments to evaluateGoogle street view images using multiple participants. We constructed a modelthat estimates the evaluation value of these images with and without the depthimages using the learning-to-rank method with deep convolutional neuralnetwork. The results demonstrate the extent to which the generalizationperformance of the streetscape evaluation model changes depending on thepresence or absence of depth images.

In this study, we developed a method for generating omnidirectional depth imagesfrom corresponding omnidirectional RGB images of streetscapes by learningeach pair of omnidirectional RGB and depth images created by computergraphics using pix2pix. Then, the models trained with different series of imagesshot under different site and weather conditions were applied to Google streetview images to generate depth images. The validity of the generated depth imageswas then evaluated visually. In addition, we conducted experiments to evaluateGoogle street view images using multiple participants. We constructed a modelthat estimates the evaluation value of these images with and without the depthimages using the learning-to-rank method with deep convolutional neuralnetwork. The results demonstrate the extent to which the generalizationperformance of the streetscape evaluation model changes depending on thepresence or absence of depth images.

Palavras-chave:

DOI: 10.5151/proceedings-ecaadesigradi2019_339

Referências bibliográficas

• [1]

Como citar:

Kinugawa, Hina; Takizawa, Atsushi; "Deep Learning Model for Predicting Preference of Space by Estimating the Depth Information of Space using Omnidirectional Images", p-61-68. In: Proceedings of 37 eCAADe and XXIII SIGraDi Joint Conference, “Architecture in the Age of the 4Th Industrial Revolution”, Porto 2019, Sousa, José Pedro; Henriques, Gonçalo Castro; Xavier, João Pedro (eds.). São Paulo: Blucher, 2019.
ISSN 23186968, DOI 10.5151/proceedings-ecaadesigradi2019_339