rsangers/explainable_perception

Explainability of Deep Learning models for Urban Space Perception

Python code for the paper: Explainability of Deep Learning models for Urban Space Perception.

Setup and preprocessing

This repository has been tested on Windows with Python 3.8.

The dependencies can be installed using: pip install -r requirements.txt

This project uses the Place Pulse 2.0 dataset from Dubey et al. [1]. To use this dataset, place all images in a folder called placepulse/ in the root directory, as well as the votes.csv file containing all voting data. The code of this repository builds upon the work of A.C. Vidal [2].

To preprocess the dataset, it is advised to run the following scripts:

python image_crop.py
python place_pulse_clean.py
python placepulse_split.py

Training

Training of the models can be started by running the train.py file. Parameters such as the model to use and the attribute to train on can be specified by passing command line arguments. For more information, please run:

python train.py -h

Analysis and explainability methods

Pretrained models can be found in the models/ directory. These can be evaluated using the run_cam.py script. Parameters such as the explainability method to use can be specified at the end of this file.

References

[1] Dubey, Abhimanyu, et al. "Deep learning the city: Quantifying urban perception at a global scale." European conference on computer vision. Springer, Cham, 2016.

[2] Vidal, Andrés Cádiz. Deep Neural Network Models with Explainable Components for Urban Space Perception. Diss. Pontificia Universidad Catolica de Chile (Chile), 2021.