face-recognition-system

Interview Task - Face Recognition System

Requirements

First, Install requirementss

pip install -r requirements

How to run

To run face recognition task, run the below code in Terminal:

main.py [-h] --video_path VIDEO_PATH --database_path DATABASE_PATH [--detector_backend DETECTOR]                             
               [--recongition_detector_backend DETECTOR] [--distance_metric METRIC]
               [--recognition_model MODEL] [--threshold THRESHOLD]

detector : [ 'opencv', 'ssd', 'dlib', 'mtcnn', 'fastmtcnn', 'retinaface', 'mediapipe', 'yolov8', 'yunet', 'centerface', ]

metric = ["cosine", "euclidean", "euclidean_l2"]

model = [ "VGG-Face", "Facenet", "Facenet512", "OpenFace", "DeepFace", "DeepID", "ArcFace", "Dlib", "SFace", "GhostFaceNet", ]

Example

python .\main.py --video_path "E:\Work\DideNegarHooshNo\AI_Task\AI_Task\files\task-video.mp4" --database_path "E:\Work\DideNegarHooshNo\AI_Task\AI_Task\files\Dataset" --detector_backend "yolov8" --distance_metric cosine --recognition_model "DeepID" --threshold 1 --recongition_detector_backend opencv

Face detection

Benchmark on Laptop CPU: Intel Core i7-12650H, GPU : NVIDIA RTX 3070

Algorithm	Speed `each frame(s)/FPS`	Accuracy `mean detection in each frame`
OpenCV	0.082/12.19	1.1
SSD	0.026/38.46	0
Dlib	0.32/3.12	0.52
MTCNN	0.75/1.33	2.25
FastMTCNN	0.113/8.85	2.45
RetinaFace	4.8/0.2	5
MediaPipe	0.0094/106.38	0
YoloV8	0.043/23.25	3.8
YuNet	0.0177/56.49	0.06
CenterFace	0.25/4	1.05

Although mean detection in each frame isn't a good metric to evaluate detection accuracy, But considering that there aren't any groud truth for this video we can use this for algorithm camparision assuming that all of the detections are correct.

It's obvious that YoloV8 has a good trade-off between speed and accuracy!