improved graph convolutional neural network for skeleton action recognition

What is it about?

skeleton action recognition method based graph convolution can calculate the motion relationship between skeleton joints in non-Euclidean space. To improve the modeling capability for the skeleton graph convolution, we propose the channel attention and multi-scale graph convolution network. This network can improve the recognition accuracy on the NTU60 and NTU120 dataset with a small number of parameters introduced.

Featured Image

Why is it important?

Currently, the high-precision skeleton action recognition methods all endure a huge amount of calculation. Our method can achieve competitive accuracy with light calculation.

Perspectives