/Gaze-Data-Poisoning

Primary LanguagePythonMIT LicenseMIT

Gaze-Data-Poisoning

This project is adapted from https://github.com/hysts/pytorch_mpiigaze.

The project adds major modifications and updates to the existing method.

Modifications: Performs Adversarial poisoning (FGSM & PGD) on the fly in the training batches for the MPIIFaceGaze dataset. Performs adversarial training as a defense mechanism.