age-and-gender 2.0.0

Age and gender estimation from face images.
Pure Python install. No compiler, no model downloads.

Credits

This package uses pretrained models from dlib-models by Davis E. King. The age predictor and gender classifier were contributed to dlib-models by Cydral Technology. We are grateful to them for making these models freely available.

Installation

pip install age-and-gender

Requires Python 3.11+. Pretrained models are bundled with the package.

Quick start

from PIL import Image
from age_and_gender import AgeAndGender

predictor = AgeAndGender()
image = Image.open("photo.jpg").convert("RGB")

print(predictor.predict(image))

[{'gender': {'value': 'female', 'confidence': 100},
  'age': {'value': 26, 'confidence': 84},
  'face': [419, 266, 506, 352]},
 ...]

_{© Bill Gates family}

Using any face detector

The built-in detector is optional. Crop faces with any detector (OpenCV, MediaPipe, RetinaFace, face_recognition, etc.) and pass each crop to predict_face():

from PIL import Image
from age_and_gender import AgeAndGender

predictor = AgeAndGender()
image = Image.open("photo.jpg").convert("RGB")

for left, top, right, bottom in detect_faces(image):  # your detector
    face = image.crop((left, top, right, bottom))  # cropped face only
    print(predictor.predict_face(face))

{'gender': {'value': 'female', 'confidence': 100}, 'age': {'value': 25, 'confidence': 80}}

If you only need one attribute, gender() and age() each run a single model:

predictor.gender(face)  # {'value': 'female', 'confidence': 100}
predictor.age(face)     # {'value': 25, 'confidence': 80}

face can also be an RGB NumPy array. Convert OpenCV (BGR) frames first with cv2.cvtColor(frame, cv2.COLOR_BGR2RGB).

API

• image / face: RGB PIL.Image or [H, W, 3] uint8 NumPy array.
• face_bounding_boxes: optional (top, right, bottom, left) boxes. When given, detection is skipped.
• face in results: [left, top, right, bottom].
• confidence: integer percentage.

Models

The pretrained models come from davisking/dlib-models and are bundled with the package, so nothing needs to be downloaded.

dnn_age_predictor_v1.dat and dnn_gender_classifier_v1.dat were converted to ONNX so they run on ONNX Runtime and are easier to use from Python. The weights are unchanged, and the outputs match the original dlib networks.

What's new in 2.0

• No legacy C++ code. The C++ extension and the vendored dlib sources from 1.x have been removed. Nothing is compiled during installation.
• ONNX inference. Age and gender run on ONNX Runtime. Face detection and landmarks use the prebuilt dlib-bin wheel.
• Bundled models. AgeAndGender() works immediately. The load_* calls are no longer needed.
• Any face detector. predict_face(), gender() and age() accept cropped faces.

License

The package is released under the MIT License. The bundled models are dedicated to the public domain under CC0 1.0.