Skip to main content
Fig. 8 | Plant Methods

Fig. 8

From: Deep learning based high-throughput phenotyping of chalkiness in rice exposed to high night temperature

Fig. 8

Sources of errors for the Grad-CAM models. Images (a–d) correspond to polished rice, while image (e) corresponds to unpolished rice. The sources of error can be summarized as: a Inconsistencies in the way chalkiness is manually annotated, due to the white gradient nature of chalkiness; b Scratches or marks (referred as noise) on the chalkiness area can be interpreted as non-chalkiness; c Irregular chalkiness shape makes it hard to annotate chalkiness very precisely; d Abrasion stains can be recognized as chalkiness (white dots on the right in the figure); e Irregular shape and fuzzy boundaries affect the ground truth annotations and the predictions in unpolished rice as well

Back to article page

Plant Methods

ISSN: 1746-4811