Table 4.
Results of the ablation study for PoseNet on the JHMDB and DeepLesion datasets.
| Model | JHMDB dataset | DeepLesion dataset | ||||||
|---|---|---|---|---|---|---|---|---|
| Accuracy | Recall | F1 score | AUC | Accuracy | Recall | F1 score | AUC | |
| w./o. Graph-based temporal learning | 89.02 ± 0.02 | 87.21 ± 0.03 | 86.12 ± 0.02 | 88.03 ± 0.03 | 90.11 ± 0.02 | 88.35 ± 0.03 | 87.25 ± 0.02 | 88.15 ± 0.03 |
| w./o. Multi-modal data fusion | 90.78 ± 0.03 | 89.12 ± 0.02 | 87.89 ± 0.02 | 89.33 ± 0.02 | 91.22 ± 0.03 | 89.78 ± 0.02 | 88.92 ± 0.03 | 89.67 ± 0.02 |
| w./o. Domain-constrained optimization | 91.34 ± 0.02 | 89.67 ± 0.03 | 88.76 ± 0.02 | 89.98 ± 0.02 | 92.01 ± 0.03 | 90.45 ± 0.02 | 89.67 ± 0.03 | 90.21 ± 0.02 |
| Ours | 93.78 ± 0.02 | 91.56 ± 0.02 | 90.89 ± 0.03 | 91.47 ± 0.03 | 94.23 ± 0.02 | 92.68 ± 0.02 | 91.33 ± 0.03 | 92.12 ± 0.02 |
The values in bold are the best values.