Anuj Rayamajhi, Nishant Uprety
This project aims to develop a face classification model using the LFW (Labeled Faces in the Wild) dataset. We will train a ResNet-18 model on both uncropped and cropped face images, comparing the performance of these models. The cropping of faces is done using MTCNN (Multi-task Cascaded Convolutional Networks).
The dataset is sampled from the LFW face dataset:
- Total number of images: 9,164
- Total number of classes: 1,680
- Download and preprocess the LFW dataset.
- Crop faces from images using MTCNN.
- Train a ResNet-18 model on uncropped face images.
- Train another ResNet-18 model on cropped face images.
- Benchmark the performance of both models.
- Compare accuracy, precision, recall, and F1-score.
- Python 3.7+
- PyTorch
- torchvision
- MTCNN
- numpy
- pandas
- scikit-learn
Clone the repository and install the required packages:
git clone https://github.com/yourusername/face-classification-resnet18.git
cd face-classification-resnet18
pip install -r requirements.txt