Table 1. Averaged rank-1 accuracies (%) on CASIA-B, excluding identical-view cases.
| The number of original input frames (m) | 5 | 10a | 15 | 20b | 25 | 30c | |
|---|---|---|---|---|---|---|---|
| NM | Skeleton models | 95.2 | 96.1 | 96.8 | 96.5 | 96.7 | 96.8 |
| Horizontal flipping | 94.3 | 94.8 | 94.9 | 95.9 | 96.1 | 95.9 | |
| CycleGAN | 94.1 | 94.9 | 95.6 | 95.7 | 95.9 | 95.5 | |
| BG | Skeleton models | 88.6 | 90.2 | 90.4 | 90.4 | 91.0 | 91.2 |
| Horizontal flipping | 85.7 | 88.2 | 88.5 | 89.8 | 88.8 | 89.71 | |
| CycleGAN | 86.3 | 88.1 | 88.8 | 88.7 | 89.1 | 89.1 | |
| CL | Skeleton models | 64.8 | 71.2 | 73.5 | 73.6 | 76.1 | 77.2 |
| Horizontal flipping | 57.2 | 64.7 | 69.3 | 72.5 | 74.0 | 75.0 | |
| CycleGAN | 57.3 | 66.0 | 69.9 | 70.8 | 73.2 | 74.9 | |
Notes.
a
The accuracies of GaitSet feeding 10 frames are 94.0%, 86.9%, and 62.7% for NM, BG, and CL respectively.
b
The accuracies of GaitSet feeding 20 frames are 94.7%, 87.9%, and 68.3% for NM, BG, and CL respectively.
c
The accuracies of GaitSet feeding 30 frames are 95.0%, 87.2%, and 70.4% for NM, BG, and CL respectively.