explainable-geoai

Overview

This is the dataset and code for the work "Explainable GeoAI: Performance assessment of deep learning model visualization techniques to understand AI learning processes with geospatial data"

Installation and Prerequisities

This project is built upon PyTorch framework and its library Captum for model interpretability. To install both PyTorch and Captum, please check their official installation guide for more details.

Getting Start

Training a New Model

This project used VGG-16 as the demo model. A trained model is saved under models. To train a new model, you can run:

python train_net.py 

Generating Visualizations

To generate visualizations of a given image, you should specify the image name <image_name>. You can run:

python gen_figure.py --image-name <image_name>
# example:
python gen_figure.py --image-name crater_000001

After the generating process, you will have four images in the outputs folder:

• <image_name>.original.png: the original image with bounding box labels
• <image_name>.occ.png: Occlusion visualization
• <image_name>.cam.png: Grad-CAM visualization
• <image_name>.ig.png: Integrated Gradients visualization

The file names of the figures in the manuscript are:

• Figure 2: volcano_000039
• Figure 3: meander_000059
• Figure 4: crater_000083
• Figure 5: dunes_000010
• Figure 6: volcano_000055
• Figure 7: icebergtongue_000067