Deepfake-detection-project-for-CS523

Deepfake image detection project for CS523

Download dataset from: https://www.kaggle.com/xhlulu/140k-real-and-fake-faces

How it works:

1. Save the dataset under the same directory as the jupyter notebook files. Under this dataset there should be 3 different dir named train/, valid/, and test/. If you don't want to make changes to the path in the jupyter notebook file, you can name the dataset 140K-data/ and make sure the path to train/, valid/, and test/ is called ./140K-data/real_vs_fake/real-vs-fake/.

2. You can uncomment the Save Module cell and run the notebooks to train and save modules. Modules are saved under .h5 extension.

3. You can run cells below train with different modules to predict the model and get accuracy, results, and plots.

Pretrained model for Custom CNN is uploaded. Pretrained models for VGG16 and DenseNet are unable to be uploaded because they are too big.