Human Action Recognition

This code is implemented based on Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition.
And this code only supports the Florence 3D action dataset.
Please download from here at ./dataset/

Environments

python main.py

File Name	Description
`preprocess.py`	Preprocess the Florence 3D action dataset. Each frame is unified into 32 frames, separated by train / valid / test and dropped into a file.
`config.py`	Hyperparameter and data path setting
`main.py`	Execution File. Loads data and models, and performs training and testing.
`model.py`	Defines the General Graph Convolutional Network (GGCN) class. Construct an adjacency matrix with three consecutive graphs, and call several layers.
`layer.py`	Defines the graph convolution, standard convolution and classifier layer.
`metric.py`	Defines the accuracy function