Original Work

Memorizing Normality to Detect Anomaly: Memory-augmented Deep Autoencoder for Unsupervised Anomaly Detection

Author's Code : Github

Why code again

Author's code feels incomplete and buggy. It has the model implemented for video input but training, loss and data loader functions are missing.

I needed to use memAE for vector inputs hence this script with dummy data.

Hope it helps others as well.

Information on the Repo

memAE : main folder under which all the scripts are present.

• data : data has the scripts for data ingestion, dataloader specifically. Write your own pre-processing scripts here if needed.
• models : has the scripts related to model architecture.
• utils : has the loss functions. Add your own utility functions here.

run.py : script which has training code, and validation code. It should be only used as a reference point. Check for bugs, use at your own risk.

Major Requirements

• PyTorch 1.6
• Numpy 1.18.5
• Pandas 1.0.5