Table 6.
Results on a test set with inconsistent annotations from the three annotators. For each metric, results from two models are shown trained on different sets of data, the upper being trained on the inconsistent dataset and the lower being trained on the consistent dataset.
| Class | AP | AP@0.5 | AP@0.75 | AR@1 | AR@10 | AR@100 |
|---|---|---|---|---|---|---|
| All (207) | 23.7 | 42.2 | 25.8 | 23.0 | 42.5 | 47.8 |
| 28.1 | 48.1 | 34.8 | 26.9 | 42.1 | 45.5 | |
| A Leaves (107) | 29.1 | 47.3 | 33.6 | 17.4 | 51.6 | 57.8 |
| 29.2 | 41.8 | 39.6 | 17.7 | 55.7 | 61.3 | |
| NA Leaves (59) | 17.9 | 34.2 | 17.4 | 19.3 | 35.3 | 44.4 |
| 20.0 | 34.2 | 21.2 | 22.8 | 30.6 | 35.6 | |
| I Stalks (11) | 31.7 | 54.7 | 31.3 | 27.3 | 50.9 | 50.9 |
| 51.7 | 76.3 | 59.2 | 34.6 | 51.8 | 55.4 | |
| O Stalks (30) | 15.9 | 32.7 | 20.8 | 28.0 | 32.3 | 38.0 |
| 10.0 | 30.6 | 19.0 | 32.3 | 23.3 | 29.7 |