Anomaly Detection method based on CW-SSIM applied to Autoencoders

Description

This repository contains the Anomaly Detection method based on CW-SSIM applied to Autoencoders of the Andrea Bionda's Politecnico di Milano Master Thesis: Pixelwise Anomaly Detection exploiting Steerable Filters based methods

Usage

Execute the software

• ACTION: action to perform, one of: 'training' or 'evaluation'
• FILE: configuration file path

cd Anomaly_Detection_CWSSIM
python Main.py -a ACTION -f FILE 

It is possible to configure the training parameters in TrainingParameters.ini and the evaluation parameters in EvaluationParameters.ini. In order to replicate the experiments, it is possible to extract the pretrained autoencoder weigths file from weights.zip and write its path in the evaluation parameters file.

Installation

Clone and install:

git clone https://github.com/AndreaBiondaPolimi/Anomaly_Detection_CWSSIM.git
cd Anomaly_Detection_CWSSIM
pip install -r requirements.txt

Requirements

• opencv-python==4.1.1.26
• numpy>=1.16.4
• scipy==1.4.1
• matplotlib==3.1.1
• tensorflow-gpu==2.1.0
• scikit-image>=0.16.2
• scikit-learn==0.21.3
• albumentations==0.4.5