Sub-activity Labeling on CAD-120
86Micro Precision/Recallfull model
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| full modelTemporal segmentation=Ground-truth2012.10 | 86 | 84.2 | 76.9 | |
| no object interactionsTemporal segmentation=Ground-truth2012.10 | 85.3 | 79.6 | 74.6 | |
| full model with trackingTemporal segmentation=Ground-truth2012.10 | 82.5 | 72.9 | 70.5 | |
| no temporal interactionsTemporal segmentation=Ground-truth2012.10 | 76 | 74.5 | 66.7 | |
| sub-activity onlyTemporal segmentation=Ground-truth2012.10 | 71.9 | 60.9 | 51.9 | |
| full, multi-seg learningTemporal segmentation=None (Estimated)2012.10 | 68.2 | 71.1 | 62.2 | |
| full, 1 segment. (best)Temporal segmentation=None (Estimated)2012.10 | 66.6 | 62 | 60.8 | |
| full, 1 segment. (averaged)Temporal segmentation=None (Estimated)2012.10 | 64.3 | 63.8 | 59.1 | |
| full, multi-seg learning + trackingTemporal segmentation=None (Estimated)2012.10 | 63.4 | 65.3 | 54 | |
| SVM multiclassTemporal segmentation=Ground-truth2012.10 | 58 | 47 | 41.6 | |
| image onlyTemporal segmentation=Ground-truth2012.10 | 56.2 | 39.6 | 41 | |
| max classTemporal segmentation=Ground-truth2012.10 | 29.2 | 29.2 | 10 |