Skip to content

Deep Learning Models for Face Detection/Recognition/Alignments, implemented in Tensorflow

License

Notifications You must be signed in to change notification settings

zziawan/deepface

 
 

Repository files navigation

deepface

blackpink with deepface(vgg model)

Deep Learning Models for Face Detection/Recognition/Alignments, implemented in Tensorflow.

It is being implemented...

Main Contributers

@ildoonet @kimdwkimdw @hahahahannie @jaehobang

Models

Baseline

A baseline model use dlib face detection module to crop rois. Then they will be forwarded at the vggface network to extract features.

Experiments

LFW Dataset

Detector Recognition Model Description Set 1-EER Accuracy
VGG Paper, No Embedding, Trained Test 0.9673
VGG Paper, Embedding, Trained Test 0.9913
dlib VGG no embedding, no training on lfw Test 0.9400 0.936
dlib VGG2-Resnet no training on lfw Test 0.9680 0.949
SSD
Mobilenet-v2
VGG2-Resnet no training on lfw Test 0.990 0.973

Install

Requirements

  • tensorflow >= 1.8.0
  • opencv >= 3.4.1

Install & Download Models

$ pip install -r requirements.txt
$ cd detectors/dlib
$ bash download.sh
$ cd ../../recognizers/vggface/
$ bash download.sh
$ cd ../resnet/
$ bash download.sh

Run

Test on samples

$ python bin/run_example.py run --source_path=./samples/faces --db_path=./sample_db.pkl --img_path=./samples/blackpink/blackpink1.jpg --method=vgg2

Test on a image

$ python bin/face.py run --visualize=true --image=./samples/blackpink/blackpink1.jpg

Reference

Models

[1] VGG Face : http://www.robots.ox.ac.uk/~vgg/software/vgg_face/

[2] VGG Face in Tensorflow : https://github.com/ZZUTK/Tensorflow-VGG-face

[3] DLib : https://github.com/davisking/dlib

[4] Dlib Guide Blog : https://www.pyimagesearch.com/2017/04/03/facial-landmarks-dlib-opencv-python/

[5] VGG Face 2 Project : https://www.robots.ox.ac.uk/~vgg/data/vgg_face2/

[6] Kera VGG Face2 : https://github.com/rcmalli/keras-vggface

Datasets

[1] LFW : http://vis-www.cs.umass.edu/lfw/

About

Deep Learning Models for Face Detection/Recognition/Alignments, implemented in Tensorflow

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 97.2%
  • Shell 2.8%