Table 4 Ablation of classification accuracies with SW-TT: Evaluating the impact of SW-TT vs. MS-TC, and the effects of omitting depthwise convolution, parting temporal window operation or shifting window operation from SW-TT.
From: Hierarchical intertwined graph representation learning for skeleton-based action recognition
Methods | X-Sub (%) | X-Set (%) |
|---|---|---|
IC-GC + MS-TC | 86.4 | 87.5 |
SW-TT w/o Dw Conv. | 86.1 − 0.3% | 87.6 +0.1% |
SW-TT w/o PTW | 86.6Â +0.2% | 88.0Â +0.5% |
SW-TT w/o SWin | 86.7Â +0.4% | 88.4Â +0.6% |
HI-GCN | 87.2Â +0.8% | 88.6Â +1.1% |