livenessdetectionnew's Introduction

Liveness Detection

This code trains and implements via video from the pyimagesearch liveness detection blog
We modified the blog's original shallow CNN model to Resnet50 that can achieve better accuracy

Get this code:

git clone https://github.com/joytsay/livenessDetection.git
cd livenessDetection

Pre-Trained Model and Train/Test Videos:

Download from here and put in root folder (/livenessDetection)

Setup Environment:

Tested on Windows 10 mini-conda environment via

Miniconda3-latest-Windows-x86_64.exe

Tested on MacOS M2 mini-forge environment via

brew install miniforge

Install Code:

Option 1: Auto run bat files

run 01_XXX.bat files (01~05) sequentially:

01_install.bat
02_gather.bat
03_trainLiveness.bat
04_runLiveness.bat
05_webcam.bat

Option 2: Cmd Line

conda create -n liveness python=3.8
conda activate liveness
pip install -r requirements.txt

# data pre-process
python gather_examples.py -i ./videos/fake.mp4 -o ./dataset/fake -d ./face_detector -c 0.9 -s 1 -f 0
python gather_examples.py -i ./videos/real.mp4 -o ./dataset/real -d ./face_detector -c 0.9 -s 1 -f 0
python gather_examples.py -i ./videos/mask.mp4 -o ./dataset/mask -d ./face_detector -c 0.9 -s 1 -f 0

# train model
python train_liveness.py -d ./dataset -m liveness.model -l le.pickle

# run liveness model on test video
python liveness_demo.py -m liveness.model -l le.pickle -d ./face_detector -c 0.5
press "q" to quit

# run liveness model on web cam
python webcam.py -m liveness.model -l le.pickle -d ./face_detector -c 0.5