Image Counterfactual Generator

How to use it

This package was tested using Python 3.7

Install Requirements

pip install -r requirements.txt

Using MobileNetV2 Model

To generate a targeted counterfactual for the MobileNetV2 model, you can use the script gen_t2_mnv2, which has the mandatory fields described below:

python gen_t2_mnv2.py --data  ./chihuahua_test/ --output ./cf_region_test  --cclass "French bulldog"

Base arguments description:

• --data - The source field where the images are
• --output - The output folder where the CF images will be saved
• --cclass - The name of the counterfactual class being targeted (choose any label from https://storage.googleapis.com/download.tensorflow.org/data/ImageNetLabels.txt)

Additional arguments:

• --jobs - Number of jobs to run the experiment (DEFAULT = 1)
• --mode - Type of replacement of segmented images, options are: mean, blur, random or inpaint (DEFAULT = blur)
• --timeout - Maximum time allowed to generate a counterfactual (DEFAULT = 60)