Skip to main content
Fig. 5 | Plant Methods

Fig. 5

From: Semantic segmentation of plant roots from RGB (mini-) rhizotron images—generalisation potential and false positives of established methods and advanced deep-learning models

Fig. 5

Regression of total root length (mm) per image as derived from manually, human labelled masks and as predicted by the best models, all trained on augmented mixed data (Table 2), predicting on unseen, different Cichorium intybus (RootPainter) dataset a, b and c are U-Net models with default (UNetGNRes), SE-ResNeXt-101 and EfficientNet-b6 decoders trained on augmented data, respectively. Formulas indicate the slope and offset of linear regressions; shaded areas represent confidence interval at 95%. Models predict less root length than manually labelled masks. The 1:1 line is shown as a dashed line. R2 values indicate goodness of fit (n = 1537). See Table 3 for evaluation metrics

Back to article page

Plant Methods

ISSN: 1746-4811