Unsupervised Anomaly Detection using Deep Learning

Automated defect detection has the potential to drastically improve efficiency in manufacturing by quickly and automatically detecting defects in objects such as samples from a production line. This can be achieved through the use of diffusion models. The popularity of diffusion models can be attributed to its capacity for high quality image generation, which has created an interest in applying diffusion models for defect detection tasks. However, diffusion models tend to have long computation times. This project proposes methods for reducing the computation time for diffusion models to allow for their usage in automatically detecting and localising defects in images.