We have used very recent computer vision models based on Visual Transformers to predict for each pixel of an outdoor photo, which class of object it is, e.g. "street", "vegetation", "water, etc. This will be used in further research for investigating the effect of water bodies and vegetation on the human mind.

• Maarten Schermer

MetaSynth is a Python package that creates a new standard to share statistical information and generate synthetic data. One part of the project fits a distribution to each of the variables, while the other part can generate a synthetic dataset from this information. In between, there is a metadata file that contains the condensed information and is human-readable. There will be privacy plugin packages that ensure that the privacy of the subjects of the initial study remain guaranteed.

Exposure to natural outdoor environments including greenery is vital for people’s health. Remote sensing via satellites is most commonly used to generate measures of greenery, however this does not capture the street-level perspective that people experience. On-site visits are also biased through subjective ratings, labor-intensive, time-consuming, and inefficient at a large-scale. We will address this research problem with an assemblage of IT methods to derive measures of greenery from street view services on the web which allows virtual navigation through urban spaces composed of geo-tagged street-level images. We will develop methods to automatically compute objective and accurate measures of greenery employing big street view data and deep learning.

• dr. Marco Helbich
• dr. ir. R.W. (Ronald) Poppe
• prof. dr. Daniel Oberski