Published November 28, 2022 | Version 1.0.0
Journal article Open

Characterisation of urban environment and activity across space and time using street images and deep learning in Accra

Creators

  • 1. Imperial College London
  • 2. University of Massachusetts, Amherst
  • 3. University of Ghana
  • 4. University of British Columbia
  • 5. McGill University

Description

This repository contains the image labelling protocol, analysis code, trained object detection model, object count data and site metadata for "Characterisation of urban environment and activity across space and time using street images and deep learning in Accra", Scientific Reports (2022).

To reproduce all main figures of the paper, extract the object_detection_analysis.zip file, navigate into the directory "OD_paper_release" and run the Jupyter notebook "Data processing and plotting.ipynb"

So that results are reproducable and consistent, we have prepared a Docker image with all necessary libraries pre-installed for running this notebook out of the box. 

This can be done using the two bash commands on a unix system (installed with Docker)

> docker pull thicknavyrain/tensorflow-object-detection-api:latest

> docker run -v "$PWD":/local -w /local -p 8888:8888 -e GRANT_SUDO=yes --user root thicknavyrain/tensorflow-object-detection-api:latest jupyter-notebook --allow-root --ip=0.0.0.0 --port=8888 --no-browser

This should open up a Jupyter environment inside the docker image which can run the notebook without needing any library installations.

To use our pre-trained object detection model, extract "object_detection_model.zip" inside which there is a "scripts" directory with the file "OD_to_file.py". Run this script in Python 3 with the "-h" flag in command line for full instructions. Note that the necessary Tensorflow Object Detection API library dependencies are also installed in the same Docker image as above and therefore, running inference within a Docker container of this environment is highly recommended for correct usage!

To extend functionality across platforms, an ONNX model release is also packaged in "object_detection/models/onnx_format/"

We highly encourage use of our model, with appropriate attribution:

Nathvani, R., Clark, S.N., Muller, E. et al. Characterisation of urban environment and activity across space and time using street images and deep learning in Accra. Sci Rep 12, 20470 (2022). https://doi.org/10.1038/s41598-022-24474-1

Contact r.nathvani@imperial.ac.uk for any troubleshooting issues.

Files

Object Detection Labeling Examples.pdf

Files (2.2 GB)

Name Size Download all
md5:c96610eab90497111c25d567e76bd846
93.4 MB Preview Download
md5:3ed5db63b47b2fb0265a44590a8a4e64
20.3 MB Preview Download
md5:1acdfe521e2c1fe5e80ffb4ced63382f
1.2 GB Preview Download
md5:02c9e74742a06a9c00fe867b2f03dac9
887.9 MB Preview Download

Additional details

Related works

Is source of
Journal article: 10.1038/s41598-022-24474-1 (DOI)

References

  • Nathvani, R., Clark, S.N., Muller, E. et al. Characterisation of urban environment and activity across space and time using street images and deep learning in Accra. Sci Rep 12, 20470 (2022). https://doi.org/10.1038/s41598-022-24474-1