---

Multi-Task Learning (MTL)

We train the model to predict both age and gender simultaneously, using the following loss functions:

Task	Output Type	Loss Function
Age	Regression	MSELoss (Mean Squared Error)
Gender	Binary Classification	BCEWithLogitsLoss

---

Challenge: Loss Scale Imbalance

The age loss (MSE) often has a larger magnitude than the gender loss (BCE), which can make training unstable.

Solution: Learnable Uncertainty Weighting

We apply the technique from:

"Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics" (Kendall et al.)

Idea:

The model learns task-specific uncertainty parameters ($\sigma^2$) that automatically scale each loss:

Where:

• L_age = Age regression loss (MSE)
• L_gender = Gender classification loss (BCE)
• σ₁, σ₂ = Learned uncertainties (as log-variance)

This allows the model to dynamically balance the contributions of each task based on its uncertainty during training.

---

Run the Project

# Clone the repository
git clone https://github.com/baduyne/Age-Gender-Prediction-from-Face-Images-using-CNN.git
cd Age-Gender-Prediction-from-Face-Images-using-CNN

How to Run from Command Line You can run the model in two modes:

• Mode 0: Use Webcam:

python executing_model.py --mode 0
# or shorthand
python executing_model.py -m 0

• Mode 1: Predict from a Static Image

python executing_model.py --mode 1 --image_path path_to_image.jpg
# or shorthand
python executing_model.py -m 1 -i path_to_image.jpg

Requirements Ensure you have the necessary dependencies installed:

pip install -r requirements.txt

(づ｡◕‿‿◕｡)づ Thank you for your interest on my project.