Facial Recognition with ArcFace Models

This repository implements a facial recognition system using ArcFace-based models, focusing on training and evaluation with identity classification and comparison metrics.

Features

• ArcFace Models: Implementation of two ArcFace variants, ResNetArcFace and ResNetDreamArcFace, for robust facial recognition.
• Training Pipeline: Train models with custom loss functions, optimize using SGD, and log the training loss for comparison.
• Evaluation Metrics: Supports one-to-one and one-to-many facial identity comparison, calculating metrics such as genuine and impostor distances.
• Dataset Handling: Automatic dataset preparation, including identity-based image organization and splitting into train/test sets.

Usage

1. Training
   Run the training script using Hydra for easy configuration:

   python train.py

   The trained models are saved for future use.

2. Testing
   Evaluate model performance using one-to-one and one-to-many comparisons:

   python test.py

   Results include precision metrics and visual plots.

Dataset

• Utilizes the CelebA-HQ dataset. Images are organized by identity and split into train/test sets.