UrbanFormer

Exploring the PlacePulse 2.0 dataset and explaining the urban safety perception through visual features. Paper.

Requirements

• Python>=3.12

Installation

  pip install -r requirements.txt

Data

Obtain the Place Pulse 2.0 dataset here.

Data Preparation

• Download images and pp2_raw_data.zip.

• Create a .env file, and add the path of the data downloaded and models.

    DATA_PATH=/path_to/datasets/
    MODEL_PATH=/path_to/models/
  

• Train models running notebooks/Models/Visual_Branch.ipynb.
  or notebooks/Models/Semantic_Branch.ipynb.
  or notebooks/Models/UrbanFormer.ipynb.

• Run visual explanations notebooks/Explanations/XAI_Visual.ipynb.
  Then, run semantic explanations notebooks/Explanations/XAI_Semantic.ipynb.
  Finally, perform IoU importance notebooks/Explanations/XAI_UrbanFormer.ipynb.

Citation

@inproceedings{moreno2024MhatMakes,
    author={Moreno-Vera, Felipe and Brandoli, Bruno and Poco, Jorge},
    title={What Makes a Place Feel Safe? Analyzing Street View Images to Identify Relevant Visual Elements}, 
    booktitle={2024 IEEE/WIC International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT)}, 
    year={2024},
    pages={395-399},
    doi={10.1109/WI-IAT62293.2024.00062}
}

Contact us

For any issue please kindly email to felipe [dot] moreno [at] fgv [dot] br