README.md

---
license: cc-by-sa-4.0
task_categories:
- image-classification
- image-segmentation
- image-feature-extraction
language:
- en
tags:
- street view imagery
- open data
- data fusion
- urban analytics
- GeoAI
- volunteered geographic information
- machine learning
- spatial data infrastructure
- geospatial
size_categories:
- 1M<n<10M
---

# Global Streetscapes

Repository for the tabular portion of the [Global Streetscapes dataset](https://ual.sg/project/global-streetscapes/) by the [Urban Analytics Lab (UAL)](https://ual.sg/) at the National University of Singapore (NUS).

Content breakdown:
* `data/` (37 GB)
  * 21 `csv` files with 346 unique features in total and 10 million rows each to characterise the 10 million street-level images in our dataset
* `manual_labels/` (23 GB)
  * `train/`
    * 8 `csv` files of manual labels for training computer vision models to classify 8 different [contextual characteristics](https://github.com/ualsg/global-streetscapes?tab=readme-ov-file#manually-labelled-subset-for-benchmarking) of a street view image, along with other metadata such as the image's location, city, file path etc. 
  * `test/`
    * 8 `csv` files of manual labels for model testing, along with other metadata such as the image's location, city, file path etc.
  * `img/`
    * 7 `tar.gz` files containing all images used for training and testing
* `models/` (2.8 GB)
  * 8 `ckpt` files storing the trained models
* `cities688.csv` contains basic information for the 688 cities included in the dataset, such as population, continent, image count etc.
* `info.csv` overviews the content of each `csv` file in `/data` and explains the 346 features 

This repository has a total size of about 62 GB.

Please follow this [guide](https://huggingface.co/docs/huggingface_hub/guides/download) from huggingface for download instructions. Please avoid using 'git clone' to download the repo as Git stores the files twice and will double the disk space usage to 124+ GB.

We have also provided a script `download_folder.py` to download one folder from this dataset, instead of just a single file or the entire dataset.

To download the imagery portion (10 million images, ~6TB), please follow the code and documentation in our [GitHub repo](https://github.com/ualsg/global-streetscapes).
Our [Wiki](https://github.com/ualsg/global-streetscapes/wiki/2-Download-images) contains instructions and a demo on how to filter the dataset for a subset of data of your interest and download the image files for them.

Read more about this project on [its website](https://ual.sg/project/global-streetscapes/), which includes an overview of this effort together with the background, [paper](https://doi.org/10.1016/j.isprsjprs.2024.06.023), examples, and FAQ.

To cite this work, please refer to the [paper](https://doi.org/10.1016/j.isprsjprs.2024.06.023):

Hou Y, Quintana M, Khomiakov M, Yap W, Ouyang J, Ito K, Wang Z, Zhao T, Biljecki F (2024): Global Streetscapes — A comprehensive dataset of 10 million street-level images across 688 cities for urban science and analytics. ISPRS Journal of Photogrammetry and Remote Sensing 215: 216-238. doi:[10.1016/j.isprsjprs.2024.06.023](https://doi.org/10.1016/j.isprsjprs.2024.06.023)

BibTeX:
```
@article{2024_global_streetscapes,
 author = {Hou, Yujun and Quintana, Matias and Khomiakov, Maxim and Yap, Winston and Ouyang, Jiani and Ito, Koichi and Wang, Zeyu and Zhao, Tianhong and Biljecki, Filip},
 doi = {10.1016/j.isprsjprs.2024.06.023},
 journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
 pages = {216-238},
 title = {Global Streetscapes -- A comprehensive dataset of 10 million street-level images across 688 cities for urban science and analytics},
 volume = {215},
 year = {2024}
}
```

A free version (postprint / author-accepted manuscript) can be downloaded [here](https://ual.sg/publication/2024-global-streetscapes/).