Abstract: In skeleton-based action recognition, it has been a dominant paradigm to extract motion features with temporal convolution and model spatial correlations with graph convolution. However, ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results