This is a repo for our project team's Deep Learning final project. It contains two Google Colab notebooks:
-
In IrisSegentation.ipynb, we finetune a Mask-RCNN to detect irises given an image of a face and perform additional pre-processing to create our input dataset.
-
GANDetection.ipynb, we implement a Residual Attention Network (RAN) for detecting if an image is real or GAN-generated, given an iris pair from the input image.
The model is a recreation of the one described in the 2022 paper "Robust Attentive Deep Neural Network for Exposing GAN-generated Faces," available here. Our PyTorch RAN implementation is adapted from this publicly available repo.