AgeGenderRecognitionRetail : A machine learning model that identifies age and gender.

Introducing “AgeGenderRecognitionRetail,” a machine learning model available in the ailia SDK, designed by Intel. This model is capable of identifying age and gender based on input face images.

Overview of AgeGenderRecognitionRetail

AgeGenderRecognitionRetail is a machine learning model developed by Intel for identifying age and gender. It takes face images as input and outputs age and gender information.

Source：https://pixabay.com/ja/videos/%E3%82%AB%E3%83%83%E3%83%97%E3%83%AB-%E8%8B%A5%E3%81%84%E3%81%A7%E3%81%99-%E9%9B%BB%E8%A9%B1-50020/

open_model_zoo/models/intel/age-gender-recognition-retail-0013 at master ·…

Fully convolutional network for simultaneous Age/Gender recognition. The network is able to recognize age of people in…

github.com

Architecture of AgeGenderRecognitionRetail:

AgeGenderRecognitionRetail takes 62×62 pixel face images as input and outputs gender as a 2-dimensional probability vector and age as a single numerical value.

Source：https://github.com/openvinotoolkit/open_model_zoo/tree/master/models/intel/age-gender-recognition-retail-0013

Since the age is output as a value between 0 and 1.0, multiplying it by 100 allows us to calculate the age.

Source：https://github.com/openvinotoolkit/open_model_zoo/blob/master/demos/interactive_face_detection_demo/cpp/detectors.cpp

The model architecture is a simple CNN. Caffe was used for training.

AgeGenderRecognitionRetail was trained using Intel’s internal dataset consisting of 20,000 images. This model can recognize ages from 18 to 75 years old. Since children’s images are not included, it cannot identify children’s ages.

The model supports facial angles up to 45 degrees. As accuracy tends to be higher with frontal faces, it is advisable to use complementary face orientation detection models such as HopeNet.

Source：https://github.com/openvinotoolkit/open_model_zoo/tree/master/models/intel/age-gender-recognition-retail-0013

The accuracy is a 6.99 years error for age and 95.80% accuracy for gender.

Source：https://github.com/openvinotoolkit/open_model_zoo/tree/master/models/intel/age-gender-recognition-retail-0013

In addition, in the official demo application, when applied to videos, it tracks faces to smooth out results over multiple frames. Age is updated by 5% increments, and gender is calculated by summing the probability values.

Source：https://github.com/openvinotoolkit/open_model_zoo/blob/master/demos/interactive_face_detection_demo/cpp/face.cpp

Even when wearing a mask, detection with reasonable accuracy is possible as long as the face is frontal.

Source：https://pixabay.com/ja/photos/%e7%9c%8b%e8%ad%b7%e5%a9%a6-%e3%83%9e%e3%82%b9%e3%82%af%e3%82%b5%e3%83%bc%e3%82%b8%e3%82%ab%e3%83%ab%e3%83%9e%e3%82%b9%e3%82%af-4962034/

How to use AgeGenderRecognitionRetail:

To use AgeGenderRecognitionRetail, use the following command. Recognition is possible from a webcam.

$ python3 age-gender-recognition-retail.py -v 0

ailia-models/face_recognition/age-gender-recognition-retail at master · axinc-ai/ailia-models

(Image from…

github.com

To estimate the age after performing face recognition on images in the specified folder, use the following command:

$ python3 age-gender-recognition-retail.py -i faces -d

Here is an example of execution.

https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FLI79oYR3M94%3Ffeature%3Doembed&display_name=YouTube&url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DLI79oYR3M94&image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FLI79oYR3M94%2Fhqdefault.jpg&key=a19fcc184b9711e1b4764040d3dc5c07&type=text%2Fhtml&schema=youtube

Ax Corporation is developing ailia SDK, a high-speed inference engine that utilizes GPUs for cross-platform AI applications. From consulting and model creation to SDK provision, development of AI-based applications and systems, and support, Ax Corporation offers comprehensive solutions in AI. Please feel free to contact us for inquiries.