Fig. 3

Per-category precision-recall curves generated using different number of training images. a–c Are for seedling detection in the TAMU2015, UGA2015, and UGA2018 datasets, and d, e are for weed detection in the TAMU2015 and UGA2015 datasets. A perfect test would have a PR curve that passes through the upper right corner (i.e., 100% for both precision and recall)