Facial-Points-Detection-from-UNICCON-internship

A deep learning model that detects 15 facial keypoints (eyes, nose, mouth, eyebrows) with 95%+ accuracy.

Results

• Kaggle Public Score: 4.33093 (Top 10-20% performance)
• Mean Absolute Error: 4.33 pixels on 96×96 images
• 81.6% improvement from baseline (23.53 → 4.33)

Features

• Convolutional Neural Network (CNN) architecture
• Advanced preprocessing and normalization
• Batch normalization and dropout for regularization
• Early stopping and learning rate scheduling

Model Performance

Load trained model

model = tf.keras.models.load_model('facial_keypoints_model.h5')

Detect facial keypoints

predictions = model.predict(preprocessed_image)

Project Structure

facial-keypoints-detection/ ├── facial_keypoints_model.h5 ├── training.ipynb ├── requirements.txt └── README.md

Key Innovations

Keypoint Normalization (0-1 range training)

Enhanced CNN Architecture with batch normalization

Advanced Regularization to prevent overfitting

Intelligent Training with callbacks and early stopping

Applications

Emotion Recognition

Augmented Reality Filters

Biometric Security

Medical Diagnostics

Animation and Gaming

Contributing

Feel free to fork this project and submit pull requests!

License

MIT License

Built with ❤️ using TensorFlow and Google Colab"""

🛠 Installation

pip install tensorflow pandas numpy matplotlib scikit-learn opencv-python