anodetection-aemem

Abnormal events detection using autoencoder (U-Net) and memory module.

Demo

• [PED1]
• [PED2]
• [Avenue]

Dataset

• [download directly]
• [description]

How to run

Use this folder structure, copy dataset into folder ./dataset/. For example, ./dataset/ped2/.

1. Training: Use this command, and you can freely define parameters with your own settings like

python3 Train.py --dataset_type dataset_type

Example for avenue:

python3 Train.py --dataset_type avenue

Need to help? Run this command:

python3 Train.py -h

2. Evaluation

python3 Evaluate.py --dataset_type dataset_type --model_dir your_model.pth --m_items_dir your_m_items.pt

Example for avenue:

python3 Evaluate.py --dataset_type avenue --model_dir ./pre_trained_model/avenue_prediction_model.pth --m_items_dir ./pre_trained_model/avenue_prediction_keys.pt

Need to help? Run this command:

python3 Evaluate.py -h

2. Run demo app

python3 app.py --method pred --dataset_type dataset_type

Example for avenue:

python3 app.py --method pred --dataset_type avenue

Model

pre-trained model

Fully pre-trained

• [fully pre-trained]: Please download and place it into root folder of project

Prediction hightest

• [ped1]
• [ped2]
• [avenue]

Paper pre-trained model

• [ped2 - pred]
• [ped2 - recons]
• [avenue - pred]
• [avenue - recons]

Works Cited

• [Learning Memory-guided Normality for Anomaly Detection] - [Github]

Demo Images

Avenue

Ped1

Ped2