search

CUFCTL / face-recognition

A GPU-accelerated real-time face recognition system based on classical machine learning algorithms
MIT License
23 stars 11 forks source link
face-recognition machine-learning pattern-recognition

readme

Face Recognition

This repository contains the code for the face recognition system developed by the FACE creative inquiry. We are developing an accelerated, real-time recognition system based on several popular face recognition techniques.

Installation

Ubuntu (or some other Linux distribution) is the easiest and most recommended OS to use for this project. If you don't have a a Linux distribution, you have a few options: (1) set up a dual-boot, (2) set up a virtual machine, or (3) use a remote Linux machine (such as the Palmetto cluster) through SSH.

Once you have installed all dependencies listed below, you can build this project with make.

On Ubuntu, you can install several dependencies through apt-get:

sudo apt-get install cmake gcc gfortran git libopencv-dev

On Palmetto, these dependencies are available as modules:

module add cmake/3.6.1 cuda-toolkit/8.0.44 gcc/4.8.1 git opencv/2.4.9

mlearn

mlearn is a machine learning library that was spawned from this project. Clone this repo and follow the instructions in the README to build the library.

OpenCV for Python

If you want to use the face-crop tool, you will need to install the Python package for OpenCV:

sudo apt-get install python-opencv

Usage

This repository includes a face-rec executable, which performs training, testing, and real-time recognition, and several helper scripts for getting datasets, running experiments, and face cropping.

The examples below use the ORL database, which you can set up with ./scripts/get_orl.sh.

Training and Testing

Run ./face-rec without any arguments to view all of the options.

To train and test the system with PCA and a 70/30 dataset partition:

./scripts/create-sets.py -d orl -t 70 -r 30
./face-rec --train train_data --test test_data --feat pca --loglevel=3

Cross Validation

To train and test the system 5 items with PCA and 70/30 dataset partitions:

./scripts/cross-validate.py -d orl -t 70 -r 30 -i 5 -- --pca

Notice that arguments for ./face-rec must be placed after a --. You can supply any valid arguments for `./face-rec in this way:

./scripts/cross-validate.py -d orl -t 70 -r 30 -i 5 -- --pca --pca_n1=20 --loglevel=3

Real-time Face Recognition

To create a database and perform real-time recognition on a video stream:

./scripts/create-sets.py -d orl -t 100
./face-rec --train train_data --feat pca
./face-rec --stream --feat pca

face-rec will use the default video stream and perform face detection and recognition on each video frame in real time.