RAJARAJESWARI COLLEGE OF ENGINEERING
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING
PROJECT REVIEW ON
“Facial Age and Gender Prediction using Deep Learning’’
By : Under the Guidance of :
Prince R (1RR21CS118) Mr. Rajkumar R
Samarth P (1RR21CS135) Dept. of CSE
Dhruva S (1RR21CS049) RRCE, Bengaluru.
ABSTRACT
In society, gender and age have two face AI is utilized in facial age and gender detection
characteristics that are highly significant. The to infer an a person's gender and age based on
main objectives of automated categorization their visual traits. This type of technology, which
are to create intelligent systems that can can be applied to CCTV cameras, healthcare by
effectively learn and identifying faces. Deep detecting significant health hazards, may be
learning methods are frequently employed in utilized in the surveillance and security industry
most studies and are also utilized to enhance to instantly identify possible suspects or
performance. criminals by assessing their age and gender [2].
CNNs have proven to be exceptionally skilled in
These applications use biometrics, which is a variety of computer vision tasks., These are
generally utilized forsecurity, as a crucial especially appropriate when nuanced
component. Emphasizing the significance of complexities for age, gender prediction from
this technology, the paper discusses its facial images.
potential to enhance everyday life
OBJECTIVE
Accurate Gender Classification:Predict the correct gender category (typically male
or female) for a given facial image.
Precise Age Prediction:Estimate the age of an individual based on facial features,
either as a continuous variable (exact age) or within discrete age groups (e.g., 0–
10, 11–20, etc.).
Robustness and Generalization:Ensure the model performs well across different
ethnicities, age groups, and varying image quality to prevent bias and enhance
generalization.
Minimizing Model Bias.
METHODOLOGY
Methodology Algorithm
In this study, we utilized the UTKFace dataset, Data Collection & Preprocessing: Collect a
comprising more than 23,000 trimmed, centered labeled dataset with images labeled by age
face pictures including gender, ethnicity and age. and gender. Some widely used datasets
With a total of 23,708 images, we encountered include UTKFace, IMDB-WIKI, and
only six missing age labels. Adience.
We selected this dataset due to its relatively Model Selection, Training the Model,
uniform distributions across age, gender, and Evaluation, Deployment
ethnicity, as well as its representation.
Working Model Accuracy for gender prediction and MAE
Data augmentation: Perform transformations or MSE for age prediction on the test set.
like rotation, flipping, zoom, and cropping to Cross-validation to check model robustness
increase data diversity. and prevent overfitting.
. Implementation tools
Setting Up Dependencies.
Data Loading and PreprocessingUse a dataset like UTKFace or Adience, which
contains labeled images for age and gender.
Defining the Model ArchitectureHere’s a CNN architecture that shares layers for
feature extraction, with separate branches for age and gender prediction.
Compiling the ModelDefine the loss functions for both branches: binary cross-
entropy for gender classification and mean squared error for age regression.
Training the Model, Evaluating the Model, Using the Model for Prediction.
Details of the base paper
Base Paper Title: Facial Age and Key Contributors:
Gender Prediction using Deep Learning The training procedure was carried out in an effort to
Journal: IEEE Access, Date of create reliable models for Convolutional Neural
publication June 2024 Networks (CNNs)-based face age and gender detection
[15].
Authors:
• DR. KALPANA R, PROFESSOR For preserving most noticeable features, the model
• D DEEPIKA, employs MaxPooling2D layers rather than
• A KAVYA,
• P HIMABINDHU, AveragePooling2D layers after each Conv2D layer. It
does this by choosing the maximum value from each
pooling window.
