Skip to content
Back to Projects

Facial Keypoints Detection

Computer vision solution using neural network stacking to detect 15 facial landmarks, achieving 2nd place in Kaggle competition.

Winner 2nd Place (RMSE: 1.28637) UC Berkeley
View on GitHub

Skills

Computer Vision Deep Learning Keypoints Detection Model Stacking

Tools

Python Keras TensorFlow OpenCV

Developed machine learning solutions to detect up to 15 facial landmarks in grayscale facial images for a Kaggle competition.

Challenge

The competition required handling a host of data quality issues including missing or incorrect labels, degraded images, and varied facial positioning across 1,783 grayscale images.

Approach

Collaborated with William Casey King, Ph.D. (Yale University) to develop a generalized stacking approach using convolutional neural networks. The solution addressed real-world challenges in image processing and neural network design.

Technologies

Built with Python using Keras and TensorFlow for deep learning, OpenCV (cv2) for image processing, and Jupyter, Pandas, and NumPy for data manipulation and analysis.