Memory Enhanced Spatial-Temporal Graph Convolutional Autoencoder for Human-related Video Anomaly Detection
This is the implementation of the paper "Memory Enhanced Spatial-Temporal Graph Convolutional Autoencoder for Human-related Video Anomaly Detection"
- Pytoch 1.7.0
- Numpy
- SciPy
- Sklearn
ShanghaiTech
please download the data from the following link
Link: trajectories
.
├── models -- Including graph definitions and convolution operators
├── utils -- Data process and score utils
├── data -- Dataset directory
├── README.md
└── train_eval.py -- Main file for training / inference
To train a model from scratch you should look up the model's configuration options.
Here is one example:
python train_eval.py --data_dir {dataset folder} --exp_dir {path to save experiment result}