Master Thesis - Anomaly Detection in Images

Anomaly detection in images is a key computer vision task and refers to detecting images or parts of images that deviate from the norm. Recent development has significantly pushed the field forward while opening up new areas of research. Example scientific literature include:

Rippel et al: “Modeling the Distribution of Normal Data in Pre-Trained Deep Features for   Anomaly Detection”

Cohen et al: “Sub-Image Anomaly Detection with Deep Pyramid Correspondences”

Possible thesis ideas:

• Investigate various ways of discovering anomalies in image feature space
• Fine-tuning pre-trained networks on images with good samples using one-class classification

The goal is to evaluate and extend on top of recent research to gain insight into what key factors make these methods work.

Prerequisites

• Computer Vision & Machine Learning knowledge
• Programming skills

Contact

For more information, contact R&D Manager Andreas Wrangsjö, andreas.wrangsjo@sick.se

We look forward to your application before  the 18th of October!