This repository contains a variety of notebooks that I created over a six-month period, where I conducted extensive research on anomaly detection. The focus of my research was on detecting anomalies in stone structures, a project called "Pierre".
The main approach in this project was to utilize pre-trained deep learning models, particularly Variational Autoencoders (VAEs), for efficient and effective anomaly detection. The goal was to find structural anomalies that may otherwise go unnoticed by human inspectors or other conventional methods.
-
anomaly_detection_MVTEC.ipynb- This notebook contains the core anomaly detection algorithm, which is based on a pre-trained model. -
Extracting_features_Material_data.ipynb- In this notebook, I focused on extracting features from the material data that were used for the model training. -
extracting_features_EfficientNetB4.ipynb- This notebook demonstrates how I used the EfficientNetB4 model for feature extraction. -
extracting_features_EfficientNetB4_MNIST.ipynb- A variant of the above, this notebook shows the use of EfficientNetB4 with the MNIST dataset. -
transfer_learning_mnist.ipynb- This notebook showcases the use of transfer learning techniques on the MNIST dataset.
I have worked in collaboration with @thomasfloquet