mrtpk/print_attack

2D print attack detection

Data preparation

Face crops are needed to train. Cropped faces has to be converted to face embeddings and should be placed in data/image_recognition/processed and data/print_attack/processed. A sample structure is in the data directory. Cropped faces can be converted to face embeddings using convert() in convert_imgs.py. After placing the embeddings in the required folder structure, run generate_dir_meta() in convert_imgs.py for data/image_recognition/processed and data/print_attack/processed.

Training

The DL models are trainined using train_dnn.py. The ML models- SVM, GMM, and logistic regresssion can be trained using svm.py, gmm.py and log_regression.py.

The face recognizer is trained using face_recognition.py.

Inference

Real time inference can be done using app.py.

Setup:

• Install Anaconda.
• conda env create -f environment.yml to create an environment.
• conda activate print2d to activate the environment.