Table 5.
Pose estimation model errors.
| Participant | Train set error (pixels) | Holdout set error (pixels) |
|---|---|---|
| P01 | 1.45 | 4.27 |
| P02 | 1.44 | 3.58 |
| P03 | 1.58 | 6.95 |
| P04 | 1.63 | 6.02 |
| P05 | 1.43 | 3.42 |
| P06 | 1.43 | 6.63 |
| P07 | 1.51 | 5.45 |
| P08 | 1.84 | 10.35 |
| P09 | 1.40 | 4.05 |
| P10 | 1.48 | 7.59 |
| P11 | 1.51 | 5.45 |
| P12 | 1.52 | 4.73 |
Root-mean-square reconstruction errors of our automated markerless pose models are shown for each participant’s train (950 frames) and holdout (50 frames) sets. We used manual annotations as ground truth when computing the error, which was averaged across all 9 upper-body keypoints. For reference3,pixels are approximately equal to 1 cm.