fawkes/evaluation

Fawkes Evaluation

We offer two ways to test the protection is effective, 1) train a local face recognition model using transfer learning, 2) use Microsoft Azure API. Note that we can't guarantee the protection is always successful due to new development in face recognition technique.

Evaluation with Local Model

To evaluate using local model, you are highly recommended to have a GPU device and train the model on it. Otherwise, the evaluation will be extremely slow and might even damage the CPUs on some machine.

To evaluate, run python3 eval_local.py -d IMAGE_DIR. Where IMAGE_DIR is the image directory send to Fawkes protection code, and it must contain both original images (testing) and cloaked image (training). All images in the directory must belong to the same person and have at least 10 images in them. Also, you cannot turn on --separate-target during the protection. (We are working on reducing some of these limitations.) The script will output the protection success rate at the end.

Evaluation with Microsoft Azure

forthcoming...