Sensitivity of deep learning facial recognition models to adversarial inputs

Fall 2018-2019

• noise.py contains our scripts for adding noise to images
• merge_training_set.py is used to merge noisy and non-noisy inputs for adversarial training
• limit_training_set.py is used to limit the number of training images used to balance images per class
• facial_landmarks/landmarks_nn.py stores our DNN used to perturb facial landmarks; it can be trained on a laptop