Sign In

Communications of the ACM

ACM TechNews

Social Media, AI Can Measure Aesthetic Quality of Landscapes

View as: Print Mobile App Share: Send by email Share on reddit Share on StumbleUpon Share on Hacker News Share on Tweeter Share on Facebook
Assessing the aesthetic quality of a landscape image.

To develop their model, the scientists trained a deep-learning algorithm on over 200,000 photos of landscapes in Great Britain obtained from the Scenic-Or-Not database.

Credit: iStock

Researchers at Switzerland's École polytechnique fédérale de Lausanne (EPFL) and the Netherlands' Wageningen University used deep learning to develop a model that can factor people's aesthetic enjoyment of a landscape into ecosystem service assessments.

The researchers trained the algorithm on more than 200,000 photos of U.K. landscapes from the Scenic-Or-Not database, which were rated via crowdsourcing as to their aesthetic quality.

The model then was run on more than 9 million photos of U.K. landscapes posted on Flickr to predict scenic beauty.

The researchers found that at a resolution of 500 square meters, the Flickr model is more accurate than the conventional environmental indicator-based model.

EPFL's Devis Tuia said, "This indicates that the use of social media-based data provides a combination of information about the state of the environment and how people interact with it. Such information has never before been obtained with such a high degree of accuracy."

From EPFL News (Switzerland)
View Full Article


Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA


No entries found