Face Recognition with Pytorch (CNN and TripletLoss)

Project Goal

The goal of this project is to build a Convolutional Neural Network (CNN) to generate facial embeddings using triplet loss. The model will be trained to create embeddings in such a way that images of the same individual are positioned closer together in embedding space, and images of different individuals are further apart.

Using the CelebA dataset, I aim to identify faces by comparing embeddings and calculating Euclidean distances between them.

Project Overview

1. Model Architecture: FaceEmbeddingCNN
2. Training Process with Triplet Loss
3. Data Preparation to Test Model
4. Embedding Generation for Celebrity Dataset
5. Face Recognition Testing
6. Evaluation and Calculation

Project in Progress ...

I intend to further develop the project, particularly focusing on improving the model's results by adjusting hyperparameters and modifying the CNN architecture. Through this project, I'm deepening my knowledge of CNNs and Pytorch.