The following files are intended to be used directly, to test the provided sample notebook. This quickstart and example
- run_face_verify.py – python script to compare faces in two images and returns the match confidence score for the images, using the Face SDK for Cognitive Services.
- analyze_face_verify_fairness.ipynb– Jupyter Notebook to generate:
- Fairness comparison table for subgroup with different evaluation metrics: True Positive Rate (Recall), False Positive Rate, False Negative Rate.
- Interactive fairness dashboard
- face_verify_sample_rand_data.csv – sample data representing results generated from the verify run on fairness dataset
The following files are intended to be replaced by the user. They are provided to illustrate how run_face_verify.py works.
- golden_labels.csv - ground truth for testing data (follow data format for your dataset)
- testing_data folder: testing data including two persons with two images/person.
- Install Anaconda
- In Anaconda run: pip install azure-cognitiveservices-vision-face to install Face SDK
-
Run face verify API
- Replace the Subscription Key and regional Base URL in “run_face_verify.py” under “face_verify_demo" directory
- In Anaconda, run “python run_face_verify.py" under “face_verify_demo" directory
- Matching results “face_verify_testing_data.csv” will be generated based on “golden_labels.csv” and images in “testing_data” folder
Note: All sample files are provided as examples. Users should replace them with their own testing data, and a corresponding golden_labels CSV files that point to their image file paths.
-
Run fairness analysis based on face verify results
- Replace " face_verify_sample_rand_data.csv" in "analyze_face_verify_fairness.ipynb" under “face_verify_demo" directory with the data you generated from "run_face_verify.py" script. You can also use the default file to show complete fairness analysis on our provided sample data.
- In Anaconda, run “jupyter notebook”
- In the pop-up browser, open “analyze_face_verify_fairness.ipynb" under “face_verify_demo " directory
- Click “Run” button to run each cell or “>>” button to run the whole notebook.