Physical Disorder of Streets in 264 Chinese Cities (2015)

Type & Source

This dataset and accompanying Python code replicate the analysis described in the paper “Measuring physical disorder in urban street spaces: A large-scale analysis using street view images and deep learning.” It involves a nationwide assessment of physical disorder using Street View imagery and a deep learning pipeline based on MobileNetV3.

Contents

1. Python Code (MobileNetV3):
2. A PyTorch implementation of MobileNetV3 trained to detect 15 disorder-related visual features (e.g. abandoned buildings, graffiti). Pre-trained models and usage instructions are included (README.md).
3. Training Samples:
4. Labeled image samples used to train the model, also suitable for retraining or transfer learning.
5. Virtual Audit Tool:
6. A lightweight Python GUI tool for image labeling. Supports sorting images into folders by label.
7. Physical Disorder Results (Shapefiles):
8. china_cities.shp
9. Attributes:
10. ID: unique city ID
11. name, nameEng: city names in Chinese and English
12. mode: spatial pattern of disorder — (a) scattered, (b) diffused, (c) linear
13. pdvalue: composite physical disorder score
14. Disorder values per factor (e.g. AB = abandoned buildings)
15. china_streets.shp
16. Attributes:
17. streets, name, nameEng
18. pdvalue: street-level disorder score
19. Factor-specific values (e.g. TR = trash)
20. china_svipoints.shp
21. Attributes:
22. point_id, street_id
23. pdvalue: point-level disorder score
24. Factor-specific values for ~1.2 million sample points

Access & Document

Physical disorder of streets in 264 Chinese cities (2015) - Mendeley Data

Updated Protocol, Code & Model

The original nationwide dataset remains available as Mendeley Data version 1. A later protocol release provides updated assessment code and YOLOv8 models: Mendeley Data, version 2 · DOI: 10.17632/d3d4h5bvss.2.

Citation

Chen, J., Chen, L., Li, Y., Zhang, W., & Long, Y. (2022). Measuring physical disorder in urban street spaces: A large-scale analysis using street view images and deep learning. Annals of the American Association of Geographers. https://doi.org/10.1080/24694452.2022.2114417.