Table 8.
Comparison of the proposed RGB-PoseTransformer3D to the state-of-the-arts. Notation of the header: RGB: R, depth: D, Skeleton: S, pose: P, and pose indicate joint-based heatmaps.
| Method | Modality | NTU RGB+D 60 | NTU RGB+D 120 |
|---|---|---|---|
| C-Sub/C-View (%) | X-Sub/X-Set (%) | ||
| SC-ConvNets [45] | R + D | 89.4/91.2 | 86.9/87.7 |
| SFMCB [46] | R + D | 91.3/92.8 | 89.3/90.3 |
| VPN [47] | R + S | 95.5/98.0 | 86.3/87.8 |
| 3DA [48] | R + S | 94.3/97.9 | 90.5/91.4 |
| STAR-Transformer [49] | R + S | 92.0/96.5 | 90.3/92.7 |
| HAC [50] | R + S | 95.7/98.8 | 93.7/94.5 |
| MMNet [51] | R + S | 96.0/98.8 | 92.9/94.4 |
| PoseConv3D [6] | R + P | 97.0/99.6 | 95.3/96.4 |
| Ours | R + P | 97.2/99.6 | 95.4/96.6 |
| Ours+PoseConv3D | R + P | 97.3/99.6 | 95.5/96.6 |