quekai / ATF-UAD

ATF-UAD

This repository supplements our paper "An Adversarial Time-Frequency Reconstruction Network for Unsupervised Anomaly Detection"

Published in Neural Networks 2023

Installation

This code needs Python-3.7 and pytorch 1.8.1 or higher.

pip3 install -r requirements.txt

Dataset Preprocessing

We have preprocessed all datasets and the link of them is shown as following, meanwhile we offer the checkpoints of all dataset to help you reproduce the results.

https://drive.google.com/file/d/1C3H9M0NdR3DViljjPzK889n6_FEHb4qr/view?usp=share_link
https://drive.google.com/file/d/19jNOoMbLSzAJjbUBrCE6V39oUFep0mkL/view?usp=share_link

After download the zip files, extract them to the corresponding folders.

Result Reproduction

To run a model on a dataset, run the following command:

python3 main.py --model <model> --dataset <dataset> --<process>

where <model> can be 'ATF_UAD' or other baselines. <dataset> can be one of 'SMAP', 'PSM', 'SWaT', 'WADI', 'SMD', 'MSDS', 'MBA', 'UCR' and 'NAB. <process> can be 'test' to reproduce the result based on the checkpoints and 'retrain' to retrain the models. For example:

python3 main.py --model ATF_UAD --dataset NAB --test