Paid Thesis Project: Visual Anomaly Detection on Steel Band Surfaces
This thesis project offers an exciting opportunity for students interested in computer vision and machine learning applications in industrial settings. The focus is on developing a robust visual anomaly detection system for steel bands using the state-of-the-art AnomalyDINO method, which utilizes the Vision Transformer architecture. The project begins with a proof-of-concept phase to adapt AnomalyDINO to the specific domain of steel band surfaces, including tasks such as model adaptation and data preprocessing. Subsequently, the project will explore two key areas: domain adaptation for new application settings, and improving detection performance through continual learning. In addition, there is potential to develop features that enhance the user experience by providing detailed explanations of anomalies. This project is an entry point into real-world industrial applications of artificial intelligence.