Objectives

The aim of this project is to develop an algorithm that detects faces in images and clusters them into groups of similar faces.

Approach

1. Face Detection: Utilize the RetinaFace detector for identifying faces in images.
2. Encode the Face Images: Apply various models to encode the detected faces into feature vectors.
3. Clustering: Use the DBSCAN algorithm to cluster the encoded faces.
4. Export Faces: Export faces in each cluster into separate folders for further analysis or usage.

Face Detector

RetinaFace: Employed for accurate and efficient face detection.

Models Tested

• VGG-Face
• FaceNet512
• GhostFaceNet
• CLIP

Clustering

DBSCAN: Density-Based Spatial Clustering of Applications with Noise, used for clustering the encoded face vectors.