Table 5 Comparison of computational and model complexity of GCN-based models on the NTU-120 X-Sub joint stream dataset.

From: Hierarchical intertwined graph representation learning for skeleton-based action recognition

Context

Methods

Param.

FLOPs

X-Sub(%)

F

S

✗

✓

2s-AGCN

3.45M

3.98G

84.0

✗

✓

CTR-GCN

1.49M

1.97G

84.9

✓

✗

InfoGCN

1.57M

1.84G

85.1

✓

✓

HD-GCN

1.68M

1.77G

85.7

✓

✓

HI-GCN

1.67M

1.73G

87.2

  1. The first column outlines the types of context integrated into the model construction topology, with F denoting frame-level and S denoting sequence-level context.
Back to article page