face-classify 0.0.3

This repository is used for face classification.

Face classifier

This repository is used for face classification. It can detect:

• open-closed eyes
• sunglasses
• profile/frontal
• blurry image.

You can find the repository on github.

If you want to use the library directly, go to pypi.

Disclaimer

The classifier works better with face cropped pictures, the models have been trained using this face recognition repossitory on group pictures.

Better use them in combination or crop the face picture yourself.

This repository comes with absolutely no guarantees.

Use Examples

Closed eyes:

from face_classify import classifier

classifier(name='face1', path='face_classify/data/open_test/closed5.jpg')

{'name': 'face1',
 'blurry': 0.0,
 'profile': 0.0,
 'sunglasses': 0.002,
 'eyes': (0.009, 0.011)}

Open eyes:

classifier(name='face2', path='face_classify/data/open_test/open2.jpg')

{'name': 'face2',
 'blurry': 0.0,
 'profile': 0.0,
 'sunglasses': 0.0,
 'eyes': (0.763, 0.782)}

Profile image

classifier(name='face3', path='face_classify/data/profile_test/profile2.jpg')

{'name': 'face3',
 'blurry': 0.0,
 'profile': 1.0,
 'sunglasses': None,
 'eyes': None}

Sunglasses image

classifier(name='face4', path='face_classify/data/sunglasses_test/sunglass1.jpg')

{'name': 'face4',
 'blurry': 0.0,
 'profile': 0.093,
 'sunglasses': 0.93,
 'eyes': None}

Blurry image

classifier(name='face5', path='face_classify/data/blurry_test/blurry12.jpg')

{'name': 'face5',
 'blurry': 0.998,
 'profile': None,
 'sunglasses': None,
 'eyes': None}

Requirements

• matplotlib==3.3.4
• numpy==1.19.5
• tensorflow==2.9.1

Models

These models have been trained separately using the code from this other repository that I created for 'weddings' pictures classification. Specially bad pictures with closed eyes.

I used a pretrained model MobilenetV2. And then retrained using my own dataset.

The models have been trained with around 10k pictures each.

I cannot upload the datasets as they are mostly from weddings of a friend photographer.

How it works

The program will check conditions and then continue or not depending on the result:

1. Blurry picture
2. Profile/Frontal
3. Sunglasses
4. Open-Closed eyes