Skip to content

Latest commit

 

History

History
24 lines (14 loc) · 1.36 KB

README.md

File metadata and controls

24 lines (14 loc) · 1.36 KB
Raw

A Demo Face Recognition Application

By - Sourav Agrawal

This application classifes a face into a given set of classes. This classes are specified during training time and the model then builds a class dictionary to be used later. The application has already been trained on a sample dataset of ~ 1000 classes.

Basic Procedure for training:

  1. Create the dataset (dataset/create_dataset.py)
  2. Load the dataset (dataset/load_dataset.py)
  3. Perform training on ResNet-50(train.py)

For Testing:

  1. Perform test directly using test.py

For face extraction while building the dataset, MTCNN module from facenet-pytorch is used. The data and mtcnn inside dataset folder are directly taken without any modification.

The initial weights are taken from VGGFace2-pytoch The weights were trained on the VGGFace2 dataset on a ResNet-50 model from scratch. Initialization Weight file : resnet50_scratch

A sample use of the package is shown in google colab. Link

You can find all the necessary resources here