About The Project

Detecting and Recoginising faces have become an extremly important task in computer vision.

Dataset

We have created our own face dataset for the task of identity recoginition. Each class in the dataset contains ~600 images of one person, the class name being the name of the directory.

To download the dataset run the following command:

pip install gdown
gdown https://drive.google.com/uc?id=1bnj8sdugWK-L40JemSTNvsML3cNuq-iG

Similarly, we have exploited the pretrained models from Yolo-v5. These prtrained models can also be downloaded using the following command:

gdown https://drive.google.com/uc?id=17V1uJSLp9KheB4t9MFdROaktgSieN6CU

This will create dataset.zip and weights.zip which should be extracted in the root folder of this project.

Built With

This project was built with

• python v3.7.12
• PyTorch v1.7
• The environment used for developing this project is available at environment.yml.

Getting Started

Clone the repository into a local machine and enter the root directory of this project using:

git clone git@github.com:here-to-learn0/RealTimeFaceDetectionAndTracking.git
cd RealTimeFaceDetectionAndTracking/

Prerequisites

Create a new conda environment and install all the libraries by running the following command

conda env create -f environment.yml

Adding own dataset

The main feature of this Face Classifier is that it is adaptable to new data classes. If you want to adapt this model on your custom live dataset, you have to first do a few steps.

Adding images

First, to classify your own face you first need to add your images to the dataset. We have made this simple. You can now run the script below which will take ~600 frames.

python scripts/capture_video_4_dataset.py --name class-name

Note

1. Make sure there is just one person in the camera frame.
2. Try to look around so that the dataset captures your face from all angles

Getting embeddings

These newly detected faces have to be converted to embeddings for training the classifier. This is done by running this:

python scripts/get_embeddings.py

This automatically detects the newest class added and creates embeddings and saves it in dataset/embeddings/

Training classifier on this new dataset

For training the classifier (SVM) on this new dataset run this command:

python scripts/train_classifier.py

This makes a new classifier and stores it in the root directory of the project.Now this classifier can be used in classification of faces.

Runnning face detector and tracker

There are two ways this tracker can be used. First, where we would like to track a person with the identity. This option is given below:

python scripts/main.py --name name-of-person-to-track

Second, is the Collective Tracker where we are not interestedd in the identity of the person but we want to collectively track (keep) n number of people in the frame thus we find the mean coodinate that our slider camera system should chase. This can be run by using the command below:

python scripts/person_track_identity_agnostic.py --num_persons_to_track 2

Here the argument --num_persons_to_track references to the number of persons to track collectively