face-classify 0.0.9

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

• for face classifier refer to Try_IT_face.ipynb
• for group face classifier refer to Try_IT_group.ipynb

Closed eyes

from face_classify import classifier
classifier(image_or_path='face_classify/data/open_test/closed5.jpg')

Output:

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

with crop face

classifier(image_or_path='face_classify/data/open_test/closed5.jpg', crop_face = True)

Output:

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

Open eyes

classifier(image_or_path='face_classify/data/open_test/open2.jpg', crop_face = True)

Output:

{'name': 'face2',
 'blurry': 0.0,
 'profile': 0.0,
 'sunglasses': 0.0,
 'eyes': (0.918, 0.924)}

Profile

classifier(image_or_path='face_classify/data/profile_test/profile2.jpg', crop_face = True)

Output:

{'name': 'face3',
 'blurry': 0.001,
 'profile': 1.0,
 'sunglasses': None,
 'eyes': None}

Sunglasses

classifier(image_or_path='face_classify/data/sunglasses_test/sunglass1.jpg', crop_face = True)

Output:

{'name': 'face4',
 'blurry': 0.0,
 'profile': 0.0,
 'sunglasses': 0.313,
 'eyes': (0.0, 0.0)}

Blurry

classifier(image_or_path='face_classify/data/blurry_test/blurry12.jpg', crop_face = True)

Output:

{'name': 'face5',
 'blurry': 1.0,
 'profile': None,
 'sunglasses': None,
 'eyes': None}

Requirements

• matplotlib==3.3.4
• numpy==1.19.5
• tensorflow==2.9.1
• face_recognition==1.3.0

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 main goal is to find if a face has opened or closed eyes. For this purpose the program will check a condition and then continue to the next one depending on the result, the order is as follows:

1. Check if Blurry picture -> Continues if its not blurry
2. Profile/Frontal picture -> Continues if its frontal
3. Wearing Sunglasses -> Continue if not sunglasses detected
4. Check Open-Closed eyes