Using computer vision to simulate passengers perception of space.

Train stations are meant to be navigated efficiently. How easily recognizable spaces are within a station can help or hinder its navigability. To improve them, we first seek to understand how people perceive space.

Using computer vision, we map the distinctivness of Gare de Lyon and Gare Saint Lazare, train stations in Paris. We collected half a million geo-tagged 360° images and fed them through a model that told us how confident it was in identifying the space. For Gare Saint Lazare, humans validated the model's decision making.

Model versus Humans

Compare the decision-making of the model to humans for spaces inside Gare Saint Lazare.

Explore the images below and their corresponding spaces on the map to see how the model and humans differed in distinguishing the space. Participants pointed to the features in the images that most helped them determine the space. The resulting point image is converted into a heatmap to compare with the model's decision-making heatmap.

Test your ability to distinguish these spaces at SpaceMatch, a short image-based survey, and find out how you score.




top level

ground level

lower level



Carlo Ratti, Director
Fábio Duarte, Project Manager
Bill Cai, Researcher
Lenna Johnsen, Researcher
Qianhui Liang, Researcher
Zhoutong Wang, Researcher
Yuji Yoshimura, Post-doctoral researcher
Sarah Campbell, Visualization, Web
Louis Charron, Researcher, Designer
Irene de la Torre Arenas, Visualization
Snoweria Zhang, Designer


Laurent Papiernik, Chief Data Officer
Etienne Burdet, Smart City Officer - AREP

download press release

The material on this website can be used freely in any publication provided that:
1. It is duly credited as a project by the MIT Senseable City Lab
2. A PDF copy of the publication is sent to

additional material

Quantifying Legibility of Indoor Spaces Using Deep Convolutional Neural Networks: Case Studies in Train Stations

Zhoutong Wang, Qianhui Liang, Fabio Duarte, Fan Zhang, Louis Charron, Lenna Johnsen, Bill Cai, Carlo Ratti
Building and Environment, 2019.

The Senseable Guide to Paris 2 — Gare de Lyon

An exploration of how digital technologies can enhance passenger experience.

Trains of Data

This Senseable project, in collaboration with SNCF, shows changes in the perceptual shape of France based on how fast passengers can reach different parts of the country.

MIT Senseable City Lab SNCF

In the collaboration with