Pattern Recognition

General Info

This project is still in the works. The objective is to detect hand movement patterns by building a skeleton graph and giving the model spatial and temporal attention. This repository holds the Pytorch implementation of Construct Dynamic Graphs for Hand Gesture Recognition via Spatial-Temporal Attention by Yuxiao Chen, Long Zhao, Xi Peng, Jianbo Yuan, and Dimitris N. Metaxas. The essential concept is to first create a fully-connected graph from a hand skeleton, and then automatically learn the node properties and edges using a self-attention method that works in both spatial and temporal domains.

Output

Note: Media can be played if your browser resolution is high enough.

Below are some examples of the result using my custom dataset.