glisicstefan/age-estimation-resnet50

Age Estimation using ResNet50

A deep learning project focused on predicting human age from facial images using the UTKFace dataset. This project follows a modular software engineering approach, separating data processing, model architecture, and training logic.

🚀 Key Features

• —Transfer Learning: Leverages a pre-trained ResNet50 model for robust feature extraction.
• —Modular Architecture: Clean code structure with specialized scripts for data setup, model building, and training logic.
• —Data Augmentation: Implements RandomHorizontalFlip, RandomRotation, and ColorJitter to improve model generalization and reduce overfitting.
• —Optimized Training: Two-stage training process, including an initial run followed by fine-tuning with a reduced learning rate.

📊 Performance

The model achieves a Mean Absolute Error (MAE) of 4.67 years on the test set. This means that, on average, the model's prediction is within ~4.7 years of the actual age.

🧠 Inference Pipeline

To ensure high accuracy, the application follows a robust inference workflow:

1. 1.Face Detection: Uses OpenCV (Haar Cascades) to locate and crop the face from the input image. This removes background noise and mimics the UTKFace dataset format.
2. 2.Preprocessing: The cropped face is resized to 224x224 and normalized using ImageNet statistics.
3. 3.Age Estimation: The ResNet50 model processes the face and outputs a continuous value.
4. 4.Post-processing: The result is rounded to the nearest integer for a user-friendly experience.

💻 Web Interface

The project includes an interactive web dashboard built with Gradio.

• —Input: Upload an image or use your webcam.
• —Output: Real-time age estimation displayed as a clean label.

🛠️ Installation & Usage

1. 1.Clone the repository:

   git clone [https://github.com/glisicstefan/age-estimation-resnet50.git](https://github.com/glisicstefan/age-estimation-resnet50.git)
   cd age-estimation-resnet50