Table 2 Mask R-CNN instance segmentation accuracies for the real (non-generated) images of the test set divided into bounding box and segmentation. Overall average precision (AP), with thresholds at \(\text {IoU}=0.50\) and \(\text {IoU}=0.75\), and overall average recall (AR) are given
| Â | Bounding box | Segmentation | ||||||
|---|---|---|---|---|---|---|---|---|
AP | AP@0.50 | AP@0.75 | AR | AP | AP@0.50 | AP@0.75 | AR | |
Arabidopsis | 0.92 | 0.99 | 0.99 | 0.95 | 0.77 | 0.99 | 0.98 | 0.78 |
GrowliFlower | 0.86 | 0.96 | 0.92 | 0.88 | 0.78 | 0.97 | 0.92 | 0.82 |