Welcome to Face Trigger’s documentation!¶
Face Trigger¶
A framework for face-recognition based applications.
Introduction¶
“Face Trigger” (FT) is a framework for building applications that rely on face-recognition. It enables the development of an end-to-end pipeline for training a machine learning classifier for the purposes of identifying people’s faces.
Pipeline¶
- Obtain images representing the faces of individuals.
- Face Trigger preprocesses individual images to find a face within the image. It rejects images that are unfit for use.
- Augment the dataset
- It is recommended to obtain atleast 10 images per subject. However, if such a scenario is not possible, Face Trigger can create artificial samples from existing images to create a 10-core dataset.
- Training Phase 1: Deep Face Embeddings
- Face Trigger runs through the dataset to extract the embeddings from each image. These embeddings are generated by a deep neural network based on a similar architecture as described in FaceNet: A Unified Embedding for Face Recognition and Clustering
- Training Phase 2: Calibrated Support Vector Classifier
- We train a calibrated Support Vector Classifier on the embeddings. This classifier is capable of generating probability measures for each embedding fed to it as belonging to one of the subjects.
- Threshold
- The probability measures from the SVC is thresholded - measures less than the threshold are rejected to belong to an unknown subject.